8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
26 #ifdef FIO_HAVE_SYSLET
42 TD_DDIR_READ = 1 << 0,
43 TD_DDIR_WRITE = 1 << 1,
44 TD_DDIR_RAND = 1 << 2,
45 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
46 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
47 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
48 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
52 * Use for maintaining statistics
55 unsigned long max_val;
56 unsigned long min_val;
57 unsigned long samples;
64 * A single data sample
73 * Dynamically growing data sample log
76 unsigned long nr_samples;
77 unsigned long max_samples;
78 struct io_sample *log;
82 * When logging io actions, this matches a single sent io_u
86 struct rb_node rb_node;
87 struct list_head list;
91 struct fio_file *file;
93 unsigned long long offset;
98 unsigned int file_action;
102 #ifdef FIO_HAVE_SYSLET
104 struct syslet_uatom atom; /* the atom to submit */
105 struct syslet_uatom *head; /* head of the sequence */
106 long ret; /* syscall return value */
111 IO_U_F_FREE = 1 << 0,
112 IO_U_F_FLIGHT = 1 << 1,
122 #ifdef FIO_HAVE_LIBAIO
125 #ifdef FIO_HAVE_POSIXAIO
129 struct sg_io_hdr hdr;
131 #ifdef FIO_HAVE_SYSLET
132 struct syslet_req req;
134 #ifdef FIO_HAVE_GUASI
138 struct timeval start_time;
139 struct timeval issue_time;
142 * Allocated/set buffer and length
145 unsigned long buflen;
146 unsigned long long offset;
147 unsigned long long endpos;
150 * IO engine state, may be different from above when we get
151 * partial transfers / residual data counts
154 unsigned long xfer_buflen;
162 * io engine private data
171 struct fio_file *file;
173 struct list_head list;
176 * Callback for io completion
178 int (*end_io)(struct thread_data *, struct io_u *);
182 * io_ops->queue() return values
185 FIO_Q_COMPLETED = 0, /* completed sync */
186 FIO_Q_QUEUED = 1, /* queued, will complete async */
187 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
190 #define FIO_HDR_MAGIC 0xf00baaef
193 VERIFY_NONE = 0, /* no verification */
194 VERIFY_MD5, /* md5 sum data blocks */
195 VERIFY_CRC64, /* crc64 sum data blocks */
196 VERIFY_CRC32, /* crc32 sum data blocks */
197 VERIFY_CRC16, /* crc16 sum data blocks */
198 VERIFY_CRC7, /* crc7 sum data blocks */
199 VERIFY_SHA256, /* sha256 sum data blocks */
200 VERIFY_SHA512, /* sha512 sum data blocks */
201 VERIFY_META, /* block_num, timestamp etc. */
202 VERIFY_PATTERN, /* verify a specific pattern */
203 VERIFY_NULL, /* pretend to verify */
207 * A header structure associated with each checksummed data block. It is
208 * followed by a checksum specific header that contains the verification
211 struct verify_header {
212 unsigned int fio_magic;
214 unsigned int verify_type;
218 uint32_t md5_digest[16];
240 unsigned char thread;
241 unsigned short numberio;
242 unsigned long time_sec;
243 unsigned long time_usec;
246 struct group_run_stats {
247 unsigned long long max_run[2], min_run[2];
248 unsigned long long max_bw[2], min_bw[2];
249 unsigned long long io_kb[2];
250 unsigned long long agg[2];
254 * What type of allocation to use for io buffers
257 MEM_MALLOC = 0, /* ordinary malloc */
258 MEM_SHM, /* use shared memory segments */
259 MEM_SHMHUGE, /* use shared memory segments with huge pages */
260 MEM_MMAP, /* use anonynomous mmap */
261 MEM_MMAPHUGE, /* memory mapped huge file */
265 * The type of object we are working on
268 FIO_TYPE_FILE = 1, /* plain file */
269 FIO_TYPE_BD, /* block device */
270 FIO_TYPE_CHAR, /* character device */
271 FIO_TYPE_PIPE, /* pipe */
274 enum fio_ioengine_flags {
275 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
276 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
277 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
278 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
279 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
280 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
281 FIO_NOIO = 1 << 6, /* thread does only pseudo IO */
284 enum fio_file_flags {
285 FIO_FILE_OPEN = 1 << 0, /* file is open */
286 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
287 FIO_FILE_EXISTS = 1 << 2, /* file there */
288 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
289 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
290 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
291 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
295 * Each thread_data structure has a number of files associated with it,
296 * this structure holds state information for a single file.
299 enum fio_filetype filetype;
302 * A file may not be a file descriptor, let the io engine decide
305 unsigned long file_data;
310 * filename and possible memory mapping
314 unsigned int major, minor;
317 * size of the file, offset into file, and io size from that offset
319 unsigned long long real_file_size;
320 unsigned long long file_offset;
321 unsigned long long io_size;
323 unsigned long long last_pos;
324 unsigned long long last_completed_pos;
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;
391 struct thread_options {
399 enum td_ddir td_ddir;
400 unsigned int ddir_nr;
401 unsigned int iodepth;
402 unsigned int iodepth_low;
403 unsigned int iodepth_batch;
405 unsigned long long size;
406 unsigned long long file_size_low;
407 unsigned long long file_size_high;
408 unsigned long long start_offset;
411 unsigned int min_bs[2];
412 unsigned int max_bs[2];
414 unsigned int nr_files;
415 unsigned int open_files;
417 unsigned int odirect;
418 unsigned int invalidate_cache;
419 unsigned int create_serialize;
420 unsigned int create_fsync;
421 unsigned int end_fsync;
422 unsigned int sync_io;
424 unsigned int do_verify;
425 unsigned int verifysort;
426 unsigned int verify_interval;
427 unsigned int verify_offset;
428 unsigned int verify_pattern;
429 unsigned int verify_pattern_bytes;
430 unsigned int verify_fatal;
431 unsigned int use_thread;
433 unsigned int do_disk_util;
434 unsigned int override_sync;
435 unsigned int rand_repeatable;
436 unsigned int write_lat_log;
437 unsigned int write_bw_log;
438 unsigned int norandommap;
439 unsigned int bs_unaligned;
440 unsigned int fsync_on_close;
442 unsigned int hugepage_size;
443 unsigned int rw_min_bs;
444 unsigned int thinktime;
445 unsigned int thinktime_spin;
446 unsigned int thinktime_blocks;
447 unsigned int fsync_blocks;
448 unsigned int start_delay;
449 unsigned long long timeout;
450 unsigned int overwrite;
451 unsigned int bw_avg_time;
453 unsigned long long zone_size;
454 unsigned long long zone_skip;
455 enum fio_memtype mem_type;
457 unsigned int stonewall;
458 unsigned int new_group;
459 unsigned int numjobs;
460 os_cpu_mask_t cpumask;
461 unsigned int cpumask_set;
463 unsigned int rwmixcycle;
464 unsigned int rwmix[2];
466 unsigned int file_service_type;
467 unsigned int group_reporting;
468 unsigned int fadvise_hint;
469 unsigned int zero_buffers;
470 unsigned int time_based;
472 char *read_iolog_file;
473 char *write_iolog_file;
476 * Pre-run and post-run shell
482 unsigned int ratemin;
483 unsigned int ratecycle;
484 unsigned int rate_iops;
485 unsigned int rate_iops_min;
490 * CPU "io" cycle burner
492 unsigned int cpuload;
493 unsigned int cpucycle;
496 #define FIO_VERROR_SIZE 128
499 * This describes a single thread/process executing a fio job.
502 struct thread_options o;
503 char verror[FIO_VERROR_SIZE];
507 struct thread_stat ts;
508 struct fio_file *files;
509 unsigned int files_index;
510 unsigned int nr_open_files;
511 unsigned int nr_done_files;
512 unsigned int nr_normal_files;
514 unsigned int next_file;
515 os_random_state_t next_file_state;
521 size_t orig_buffer_size;
522 volatile int terminate;
523 volatile int runstate;
525 unsigned int last_was_sync;
535 os_random_state_t bsrange_state;
536 os_random_state_t verify_state;
541 * IO engine hooks, contains everything needed to submit an io_u
542 * to any of the available IO engines.
544 struct ioengine_ops *io_ops;
547 * Current IO depth and list of free and busy io_u's.
549 unsigned int cur_depth;
550 struct list_head io_u_freelist;
551 struct list_head io_u_busylist;
552 struct list_head io_u_requeues;
553 unsigned int io_u_queued;
558 unsigned long rate_usec_cycle;
559 long rate_pending_usleep;
560 unsigned long rate_bytes;
561 unsigned long rate_blocks;
562 struct timeval lastrate;
564 unsigned long long total_io_size;
566 unsigned long io_issues[2];
567 unsigned long long io_blocks[2];
568 unsigned long long io_bytes[2];
569 unsigned long long io_skip_bytes;
570 unsigned long long this_io_bytes[2];
571 unsigned long long zone_bytes;
572 struct fio_sem *mutex;
575 * State for random io, a bitmap of blocks done vs not done
577 os_random_state_t random_state;
579 struct timeval start; /* start of this loop */
580 struct timeval epoch; /* time job was started */
581 struct timeval rw_end[2];
582 struct timeval last_issue;
583 unsigned int rw_end_set[2];
586 * read/write mixed workload state
588 os_random_state_t rwmix_state;
589 unsigned long long rwmix_bytes;
590 struct timeval rwmix_switch;
591 enum fio_ddir rwmix_ddir;
592 unsigned int ddir_nr;
595 * IO history logs for verification. We use a tree for sorting,
596 * if we are overwriting. Otherwise just use a fifo.
598 struct rb_root io_hist_tree;
599 struct list_head io_hist_list;
604 struct list_head io_log_list;
609 struct timeval timeout_end;
610 struct itimerval timer;
613 * for fileservice, how often to switch to a new file
615 unsigned int file_service_nr;
616 unsigned int file_service_left;
617 struct fio_file *file_service_file;
620 * For generating file sizes
622 os_random_state_t file_size_state;
626 * roundrobin available files, or choose one at random.
629 FIO_FSERVICE_RANDOM = 1,
634 * when should interactive ETA output be generated
643 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
644 * the timer on each queue operation.
646 #define IO_U_TIMEOUT_INC 5
647 #define IO_U_TIMEOUT 30
649 #define __td_verror(td, err, msg, func) \
655 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
659 #define td_verror(td, err, func) \
660 __td_verror((td), (err), strerror((err)), (func))
661 #define td_vmsg(td, err, msg, func) \
662 __td_verror((td), (err), (msg), (func))
664 extern int exitall_on_terminate;
665 extern int thread_number;
666 extern int nr_process, nr_thread;
669 extern int terse_output;
672 extern int temp_stall_ts;
673 extern unsigned long long mlock_size;
674 extern unsigned long page_mask, page_size;
675 extern int read_only;
676 extern int eta_print;
678 extern struct thread_data *threads;
680 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
681 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
682 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
683 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
685 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
687 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
690 #define BLOCKS_PER_MAP (8 * sizeof(long))
691 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
692 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
693 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
695 #define MAX_JOBS (1024)
697 static inline int should_fsync(struct thread_data *td)
699 if (td->last_was_sync)
703 if (td_write(td) || td_rw(td) || td->o.override_sync)
710 * Disk utils as read in /sys/block/<dev>/stat
712 struct disk_util_stat {
715 unsigned long long sectors[2];
718 unsigned time_in_queue;
722 * Per-device disk util management
725 struct list_head list;
732 struct disk_util_stat dus;
733 struct disk_util_stat last_dus;
739 #define DISK_UTIL_MSEC (250)
750 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
751 extern void log_io_u(struct thread_data *, struct io_u *);
752 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
753 extern int __must_check init_iolog(struct thread_data *td);
754 extern void log_io_piece(struct thread_data *, struct io_u *);
755 extern void prune_io_piece_log(struct thread_data *);
756 extern void write_iolog_close(struct thread_data *);
761 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
762 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
763 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
764 extern void show_run_stats(void);
765 extern void init_disk_util(struct thread_data *);
766 extern void update_rusage_stat(struct thread_data *);
767 extern void update_io_ticks(void);
768 extern void disk_util_timer_arm(void);
769 extern void setup_log(struct io_log **);
770 extern void finish_log(struct thread_data *, struct io_log *, const char *);
771 extern void __finish_log(struct io_log *, const char *);
772 extern struct io_log *agg_io_log[2];
773 extern int write_bw_log;
774 extern void add_agg_sample(unsigned long, enum fio_ddir);
779 extern unsigned long utime_since(struct timeval *, struct timeval *);
780 extern unsigned long utime_since_now(struct timeval *);
781 extern unsigned long mtime_since(struct timeval *, struct timeval *);
782 extern unsigned long mtime_since_now(struct timeval *);
783 extern unsigned long time_since_now(struct timeval *);
784 extern unsigned long mtime_since_genesis(void);
785 extern void __usec_sleep(unsigned int);
786 extern void usec_sleep(struct thread_data *, unsigned long);
787 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
788 extern void fill_start_time(struct timeval *);
789 extern void fio_gettime(struct timeval *, void *);
790 extern void set_genesis_time(void);
793 * Init/option functions
795 extern int __must_check parse_options(int, char **);
796 extern int fio_option_parse(struct thread_data *, const char *);
797 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
798 extern void fio_fill_default_options(struct thread_data *);
799 extern int fio_show_option_help(const char *);
800 extern void fio_options_dup_and_init(struct option *);
801 extern void options_mem_dupe(struct thread_data *);
802 extern void options_mem_free(struct thread_data *);
803 #define FIO_GETOPT_JOB 0x89988998
804 #define FIO_NR_OPTIONS 128
807 * File setup/shutdown
809 extern void close_files(struct thread_data *);
810 extern int __must_check setup_files(struct thread_data *);
811 extern int __must_check open_files(struct thread_data *);
812 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
813 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
814 extern void generic_close_file(struct thread_data *, struct fio_file *);
815 extern int add_file(struct thread_data *, const char *);
816 extern void get_file(struct fio_file *);
817 extern void put_file(struct thread_data *, struct fio_file *);
818 extern int add_dir_files(struct thread_data *, const char *);
819 extern int init_random_map(struct thread_data *);
820 extern void dup_files(struct thread_data *, struct thread_data *);
821 extern int get_fileno(struct thread_data *, const char *);
822 extern void free_release_files(struct thread_data *);
827 extern void print_thread_status(void);
828 extern void print_status_init(int);
833 #ifdef FIO_HAVE_DISK_UTIL
834 extern void show_disk_util(void);
835 extern void disk_util_timer_arm(void);
836 extern void init_disk_util(struct thread_data *);
837 extern void update_io_ticks(void);
839 #define show_disk_util()
840 #define disk_util_timer_arm()
841 #define init_disk_util(td)
842 #define update_io_ticks()
846 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
847 * will never back again. It may cycle between running/verififying/fsyncing.
848 * Once the thread reaches TD_EXITED, it is just waiting for the core to
865 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
866 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
867 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
872 extern int __must_check fio_pin_memory(void);
873 extern void fio_unpin_memory(void);
874 extern int __must_check allocate_io_mem(struct thread_data *);
875 extern void free_io_mem(struct thread_data *);
880 #define queue_full(td) list_empty(&(td)->io_u_freelist)
881 extern struct io_u *__get_io_u(struct thread_data *);
882 extern struct io_u *get_io_u(struct thread_data *);
883 extern void put_io_u(struct thread_data *, struct io_u *);
884 extern void requeue_io_u(struct thread_data *, struct io_u **);
885 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
886 extern long __must_check io_u_queued_complete(struct thread_data *, int);
887 extern void io_u_queued(struct thread_data *, struct io_u *);
888 extern void io_u_log_error(struct thread_data *, struct io_u *);
889 extern void io_u_init_timeout(void);
890 extern void io_u_set_timeout(struct thread_data *);
891 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
894 * io engine entry points
896 extern int __must_check td_io_init(struct thread_data *);
897 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
898 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
899 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
900 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
901 extern int __must_check td_io_commit(struct thread_data *);
902 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
903 extern void td_io_close_file(struct thread_data *, struct fio_file *);
908 #ifdef FIO_HAVE_BLKTRACE
909 extern int is_blktrace(const char *);
910 extern int load_blktrace(struct thread_data *, const char *);
914 * If logging output to a file, stderr should go to both stderr and f_err
916 #define log_err(args...) do { \
917 fprintf(f_err, ##args); \
918 if (f_err != stderr) \
919 fprintf(stderr, ##args); \
922 #define log_info(args...) fprintf(f_out, ##args)
924 FILE *get_f_out(void);
925 FILE *get_f_err(void);
927 struct ioengine_ops {
928 struct list_head list;
932 int (*setup)(struct thread_data *);
933 int (*init)(struct thread_data *);
934 int (*prep)(struct thread_data *, struct io_u *);
935 int (*queue)(struct thread_data *, struct io_u *);
936 int (*commit)(struct thread_data *);
937 int (*getevents)(struct thread_data *, int, int, struct timespec *);
938 struct io_u *(*event)(struct thread_data *, int);
939 int (*cancel)(struct thread_data *, struct io_u *);
940 void (*cleanup)(struct thread_data *);
941 int (*open_file)(struct thread_data *, struct fio_file *);
942 void (*close_file)(struct thread_data *, struct fio_file *);
947 #define FIO_IOOPS_VERSION 7
949 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
950 extern void register_ioengine(struct ioengine_ops *);
951 extern void unregister_ioengine(struct ioengine_ops *);
952 extern void close_ioengine(struct thread_data *);
955 * Mark unused variables passed to ops functions as unused, to silence gcc
957 #define fio_unused __attribute((__unused__))
958 #define fio_init __attribute__((constructor))
959 #define fio_exit __attribute__((destructor))
961 #define for_each_td(td, i) \
962 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
963 #define for_each_file(td, f, i) \
964 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
966 #define fio_assert(td, cond) do { \
969 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
970 (td)->runstate = TD_EXITED; \
971 (td)->error = EFAULT; \
976 static inline void clear_error(struct thread_data *td)
979 td->verror[0] = '\0';