8 #include <sys/resource.h>
16 #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_CRC32, /* crc32 sum data blocks */
196 VERIFY_NULL, /* pretend to verify */
200 * A header structure associated with each checksummed data block
202 struct verify_header {
203 unsigned int fio_magic;
205 unsigned int verify_type;
207 char md5_digest[MD5_HASH_WORDS * 4];
212 struct group_run_stats {
213 unsigned long long max_run[2], min_run[2];
214 unsigned long long max_bw[2], min_bw[2];
215 unsigned long long io_kb[2];
216 unsigned long long agg[2];
220 * What type of allocation to use for io buffers
223 MEM_MALLOC = 0, /* ordinary malloc */
224 MEM_SHM, /* use shared memory segments */
225 MEM_SHMHUGE, /* use shared memory segments with huge pages */
226 MEM_MMAP, /* use anonynomous mmap */
227 MEM_MMAPHUGE, /* memory mapped huge file */
231 * The type of object we are working on
234 FIO_TYPE_FILE = 1, /* plain file */
235 FIO_TYPE_BD, /* block device */
236 FIO_TYPE_CHAR, /* character device */
237 FIO_TYPE_PIPE, /* pipe */
240 enum fio_ioengine_flags {
241 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
242 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
243 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
244 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
245 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
248 enum fio_file_flags {
249 FIO_FILE_OPEN = 1 << 0, /* file is open */
250 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
251 FIO_FILE_EXISTS = 1 << 2, /* file there */
252 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
253 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
254 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
255 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
259 * Each thread_data structure has a number of files associated with it,
260 * this structure holds state information for a single file.
263 enum fio_filetype filetype;
266 * A file may not be a file descriptor, let the io engine decide
269 unsigned long file_data;
274 * filename and possible memory mapping
278 unsigned int major, minor;
281 * size of the file, offset into file, and io size from that offset
283 unsigned long long real_file_size;
284 unsigned long long file_offset;
285 unsigned long long io_size;
287 unsigned long long last_pos;
288 unsigned long long last_completed_pos;
291 * block map for random io
293 unsigned long *file_map;
294 unsigned int num_maps;
295 unsigned int last_free_lookup;
298 enum fio_file_flags flags;
302 * How many depth levels to log
304 #define FIO_IO_U_MAP_NR 8
305 #define FIO_IO_U_LAT_U_NR 10
306 #define FIO_IO_U_LAT_M_NR 12
317 struct io_log *slat_log;
318 struct io_log *clat_log;
319 struct io_log *bw_log;
322 * bandwidth and latency stats
324 struct io_stat clat_stat[2]; /* completion latency */
325 struct io_stat slat_stat[2]; /* submission latency */
326 struct io_stat bw_stat[2]; /* bandwidth stats */
328 unsigned long long stat_io_bytes[2];
329 struct timeval stat_sample_time[2];
332 * fio system usage accounting
334 struct rusage ru_start;
335 struct rusage ru_end;
336 unsigned long usr_time;
337 unsigned long sys_time;
341 * IO depth and latency stats
343 unsigned int io_u_map[FIO_IO_U_MAP_NR];
344 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
345 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
346 unsigned long total_io_u[2];
347 unsigned long short_io_u[2];
349 unsigned long long io_bytes[2];
350 unsigned long runtime[2];
351 unsigned long total_run_time;
354 struct thread_options {
362 enum td_ddir td_ddir;
363 unsigned int ddir_nr;
364 unsigned int iodepth;
365 unsigned int iodepth_low;
366 unsigned int iodepth_batch;
368 unsigned long long size;
369 unsigned long long file_size_low;
370 unsigned long long file_size_high;
371 unsigned long long start_offset;
374 unsigned int min_bs[2];
375 unsigned int max_bs[2];
377 unsigned int nr_files;
378 unsigned int open_files;
380 unsigned int odirect;
381 unsigned int invalidate_cache;
382 unsigned int create_serialize;
383 unsigned int create_fsync;
384 unsigned int end_fsync;
385 unsigned int sync_io;
387 unsigned int verifysort;
388 unsigned int use_thread;
390 unsigned int do_disk_util;
391 unsigned int override_sync;
392 unsigned int rand_repeatable;
393 unsigned int write_lat_log;
394 unsigned int write_bw_log;
395 unsigned int norandommap;
396 unsigned int bs_unaligned;
397 unsigned int fsync_on_close;
399 unsigned int hugepage_size;
400 unsigned int rw_min_bs;
401 unsigned int thinktime;
402 unsigned int thinktime_spin;
403 unsigned int thinktime_blocks;
404 unsigned int fsync_blocks;
405 unsigned int start_delay;
406 unsigned long long timeout;
407 unsigned int overwrite;
408 unsigned int bw_avg_time;
410 unsigned long long zone_size;
411 unsigned long long zone_skip;
412 enum fio_memtype mem_type;
414 unsigned int stonewall;
415 unsigned int new_group;
416 unsigned int numjobs;
417 os_cpu_mask_t cpumask;
418 unsigned int cpumask_set;
420 unsigned int rwmixcycle;
421 unsigned int rwmix[2];
423 unsigned int file_service_type;
424 unsigned int group_reporting;
425 unsigned int fadvise_hint;
426 unsigned int zero_buffers;
427 unsigned int time_based;
429 char *read_iolog_file;
430 char *write_iolog_file;
433 * Pre-run and post-run shell
439 unsigned int ratemin;
440 unsigned int ratecycle;
441 unsigned int rate_iops;
442 unsigned int rate_iops_min;
447 * CPU "io" cycle burner
449 unsigned int cpuload;
450 unsigned int cpucycle;
453 #define FIO_VERROR_SIZE 128
456 * This describes a single thread/process executing a fio job.
459 struct thread_options o;
460 char verror[FIO_VERROR_SIZE];
464 struct thread_stat ts;
465 struct fio_file *files;
466 unsigned int files_index;
467 unsigned int nr_open_files;
468 unsigned int nr_done_files;
469 unsigned int nr_normal_files;
471 unsigned int next_file;
472 os_random_state_t next_file_state;
477 size_t orig_buffer_size;
478 volatile int terminate;
479 volatile int runstate;
481 unsigned int last_was_sync;
491 os_random_state_t bsrange_state;
492 os_random_state_t verify_state;
497 * IO engine hooks, contains everything needed to submit an io_u
498 * to any of the available IO engines.
500 struct ioengine_ops *io_ops;
503 * Current IO depth and list of free and busy io_u's.
505 unsigned int cur_depth;
506 struct list_head io_u_freelist;
507 struct list_head io_u_busylist;
508 struct list_head io_u_requeues;
509 unsigned int io_u_queued;
514 unsigned long rate_usec_cycle;
515 long rate_pending_usleep;
516 unsigned long rate_bytes;
517 unsigned long rate_blocks;
518 struct timeval lastrate;
520 unsigned long long total_io_size;
522 unsigned long io_issues[2];
523 unsigned long long io_blocks[2];
524 unsigned long long io_bytes[2];
525 unsigned long long io_skip_bytes;
526 unsigned long long this_io_bytes[2];
527 unsigned long long zone_bytes;
528 struct fio_sem *mutex;
531 * State for random io, a bitmap of blocks done vs not done
533 os_random_state_t random_state;
535 struct timeval start; /* start of this loop */
536 struct timeval epoch; /* time job was started */
537 struct timeval rw_end[2];
538 struct timeval last_issue;
539 unsigned int rw_end_set[2];
542 * read/write mixed workload state
544 os_random_state_t rwmix_state;
545 unsigned long long rwmix_bytes;
546 struct timeval rwmix_switch;
547 enum fio_ddir rwmix_ddir;
548 unsigned int ddir_nr;
551 * IO history logs for verification. We use a tree for sorting,
552 * if we are overwriting. Otherwise just use a fifo.
554 struct rb_root io_hist_tree;
555 struct list_head io_hist_list;
560 struct list_head io_log_list;
565 struct timeval timeout_end;
566 struct itimerval timer;
569 * for fileservice, how often to switch to a new file
571 unsigned int file_service_nr;
572 unsigned int file_service_left;
573 struct fio_file *file_service_file;
576 * For generating file sizes
578 os_random_state_t file_size_state;
582 * roundrobin available files, or choose one at random.
585 FIO_FSERVICE_RANDOM = 1,
590 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
591 * the timer on each queue operation.
593 #define IO_U_TIMEOUT_INC 5
594 #define IO_U_TIMEOUT 30
596 #define __td_verror(td, err, msg, func) \
602 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
606 #define td_verror(td, err, func) \
607 __td_verror((td), (err), strerror((err)), (func))
608 #define td_vmsg(td, err, msg, func) \
609 __td_verror((td), (err), (msg), (func))
611 extern int exitall_on_terminate;
612 extern int thread_number;
613 extern int nr_process, nr_thread;
616 extern int terse_output;
619 extern int temp_stall_ts;
620 extern unsigned long long mlock_size;
621 extern unsigned long page_mask, page_size;
623 extern struct thread_data *threads;
625 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
626 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
627 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
628 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
630 #define BLOCKS_PER_MAP (8 * sizeof(long))
631 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
632 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
633 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
635 #define MAX_JOBS (1024)
637 static inline int should_fsync(struct thread_data *td)
639 if (td->last_was_sync)
643 if (td_write(td) || td_rw(td) || td->o.override_sync)
650 * Disk utils as read in /sys/block/<dev>/stat
652 struct disk_util_stat {
655 unsigned long long sectors[2];
658 unsigned time_in_queue;
662 * Per-device disk util management
665 struct list_head list;
672 struct disk_util_stat dus;
673 struct disk_util_stat last_dus;
679 #define DISK_UTIL_MSEC (250)
690 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
691 extern void log_io_u(struct thread_data *, struct io_u *);
692 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
693 extern int __must_check init_iolog(struct thread_data *td);
694 extern void log_io_piece(struct thread_data *, struct io_u *);
695 extern void prune_io_piece_log(struct thread_data *);
696 extern void write_iolog_close(struct thread_data *);
701 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
702 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
703 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
704 extern void show_run_stats(void);
705 extern void init_disk_util(struct thread_data *);
706 extern void update_rusage_stat(struct thread_data *);
707 extern void update_io_ticks(void);
708 extern void disk_util_timer_arm(void);
709 extern void setup_log(struct io_log **);
710 extern void finish_log(struct thread_data *, struct io_log *, const char *);
711 extern void __finish_log(struct io_log *, const char *);
712 extern struct io_log *agg_io_log[2];
713 extern int write_bw_log;
714 extern void add_agg_sample(unsigned long, enum fio_ddir);
719 extern unsigned long utime_since(struct timeval *, struct timeval *);
720 extern unsigned long utime_since_now(struct timeval *);
721 extern unsigned long mtime_since(struct timeval *, struct timeval *);
722 extern unsigned long mtime_since_now(struct timeval *);
723 extern unsigned long time_since_now(struct timeval *);
724 extern unsigned long mtime_since_genesis(void);
725 extern void __usec_sleep(unsigned int);
726 extern void usec_sleep(struct thread_data *, unsigned long);
727 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
728 extern void fill_start_time(struct timeval *);
729 extern void fio_gettime(struct timeval *, void *);
730 extern void set_genesis_time(void);
733 * Init/option functions
735 extern int __must_check parse_options(int, char **);
736 extern int fio_option_parse(struct thread_data *, const char *);
737 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
738 extern void fio_fill_default_options(struct thread_data *);
739 extern int fio_show_option_help(const char *);
740 extern void fio_options_dup_and_init(struct option *);
741 extern void options_mem_dupe(struct thread_data *);
742 extern void options_mem_free(struct thread_data *);
743 #define FIO_GETOPT_JOB 0x89988998
744 #define FIO_NR_OPTIONS 128
747 * File setup/shutdown
749 extern void close_files(struct thread_data *);
750 extern int __must_check setup_files(struct thread_data *);
751 extern int __must_check open_files(struct thread_data *);
752 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
753 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
754 extern void generic_close_file(struct thread_data *, struct fio_file *);
755 extern int add_file(struct thread_data *, const char *);
756 extern void get_file(struct fio_file *);
757 extern void put_file(struct thread_data *, struct fio_file *);
758 extern int add_dir_files(struct thread_data *, const char *);
759 extern int init_random_map(struct thread_data *);
760 extern void dup_files(struct thread_data *, struct thread_data *);
761 extern int get_fileno(struct thread_data *, const char *);
762 extern void free_release_files(struct thread_data *);
767 extern void print_thread_status(void);
768 extern void print_status_init(int);
773 #ifdef FIO_HAVE_DISK_UTIL
774 extern void show_disk_util(void);
775 extern void disk_util_timer_arm(void);
776 extern void init_disk_util(struct thread_data *);
777 extern void update_io_ticks(void);
779 #define show_disk_util()
780 #define disk_util_timer_arm()
781 #define init_disk_util(td)
782 #define update_io_ticks()
786 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
787 * will never back again. It may cycle between running/verififying/fsyncing.
788 * Once the thread reaches TD_EXITED, it is just waiting for the core to
805 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
806 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
807 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
812 extern int __must_check fio_pin_memory(void);
813 extern void fio_unpin_memory(void);
814 extern int __must_check allocate_io_mem(struct thread_data *);
815 extern void free_io_mem(struct thread_data *);
820 #define queue_full(td) list_empty(&(td)->io_u_freelist)
821 extern struct io_u *__get_io_u(struct thread_data *);
822 extern struct io_u *get_io_u(struct thread_data *);
823 extern void put_io_u(struct thread_data *, struct io_u *);
824 extern void requeue_io_u(struct thread_data *, struct io_u **);
825 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
826 extern long __must_check io_u_queued_complete(struct thread_data *, int);
827 extern void io_u_queued(struct thread_data *, struct io_u *);
828 extern void io_u_log_error(struct thread_data *, struct io_u *);
829 extern void io_u_init_timeout(void);
830 extern void io_u_set_timeout(struct thread_data *);
831 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
834 * io engine entry points
836 extern int __must_check td_io_init(struct thread_data *);
837 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
838 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
839 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
840 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
841 extern int __must_check td_io_commit(struct thread_data *);
842 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
843 extern void td_io_close_file(struct thread_data *, struct fio_file *);
848 #ifdef FIO_HAVE_BLKTRACE
849 extern int is_blktrace(const char *);
850 extern int load_blktrace(struct thread_data *, const char *);
854 * If logging output to a file, stderr should go to both stderr and f_err
856 #define log_err(args...) do { \
857 fprintf(f_err, ##args); \
858 if (f_err != stderr) \
859 fprintf(stderr, ##args); \
862 #define log_info(args...) fprintf(f_out, ##args)
864 FILE *get_f_out(void);
865 FILE *get_f_err(void);
867 struct ioengine_ops {
868 struct list_head list;
872 int (*setup)(struct thread_data *);
873 int (*init)(struct thread_data *);
874 int (*prep)(struct thread_data *, struct io_u *);
875 int (*queue)(struct thread_data *, struct io_u *);
876 int (*commit)(struct thread_data *);
877 int (*getevents)(struct thread_data *, int, int, struct timespec *);
878 struct io_u *(*event)(struct thread_data *, int);
879 int (*cancel)(struct thread_data *, struct io_u *);
880 void (*cleanup)(struct thread_data *);
881 int (*open_file)(struct thread_data *, struct fio_file *);
882 void (*close_file)(struct thread_data *, struct fio_file *);
887 #define FIO_IOOPS_VERSION 7
889 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
890 extern void register_ioengine(struct ioengine_ops *);
891 extern void unregister_ioengine(struct ioengine_ops *);
892 extern void close_ioengine(struct thread_data *);
895 * Mark unused variables passed to ops functions as unused, to silence gcc
897 #define fio_unused __attribute((__unused__))
898 #define fio_init __attribute__((constructor))
899 #define fio_exit __attribute__((destructor))
901 #define for_each_td(td, i) \
902 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
903 #define for_each_file(td, f, i) \
904 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
906 #define fio_assert(td, cond) do { \
909 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
910 (td)->runstate = TD_EXITED; \
911 (td)->error = EFAULT; \
916 static inline void clear_error(struct thread_data *td)
919 td->verror[0] = '\0';