8 #include <sys/resource.h>
16 #include "compiler/compiler.h"
21 #include "arch/arch.h"
25 #ifdef FIO_HAVE_SYSLET
40 TD_DDIR_READ = 1 << 0,
41 TD_DDIR_WRITE = 1 << 1,
42 TD_DDIR_RAND = 1 << 2,
43 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
44 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
45 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
46 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
50 * Use for maintaining statistics
53 unsigned long max_val;
54 unsigned long min_val;
55 unsigned long samples;
62 * A single data sample
71 * Dynamically growing data sample log
74 unsigned long nr_samples;
75 unsigned long max_samples;
76 struct io_sample *log;
80 * When logging io actions, this matches a single sent io_u
84 struct rb_node rb_node;
85 struct list_head list;
87 struct fio_file *file;
88 unsigned long long offset;
93 #ifdef FIO_HAVE_SYSLET
95 struct syslet_uatom atom; /* the atom to submit */
96 struct syslet_uatom *head; /* head of the sequence */
97 long ret; /* syscall return value */
102 IO_U_F_FREE = 1 << 0,
103 IO_U_F_FLIGHT = 1 << 1,
113 #ifdef FIO_HAVE_LIBAIO
116 #ifdef FIO_HAVE_POSIXAIO
120 struct sg_io_hdr hdr;
122 #ifdef FIO_HAVE_SYSLET
123 struct syslet_req req;
125 #ifdef FIO_HAVE_GUASI
129 struct timeval start_time;
130 struct timeval issue_time;
133 * Allocated/set buffer and length
136 unsigned long buflen;
137 unsigned long long offset;
138 unsigned long long endpos;
141 * IO engine state, may be different from above when we get
142 * partial transfers / residual data counts
145 unsigned long xfer_buflen;
153 * io engine private data
162 struct fio_file *file;
164 struct list_head list;
167 * Callback for io completion
169 int (*end_io)(struct thread_data *, struct io_u *);
173 * io_ops->queue() return values
176 FIO_Q_COMPLETED = 0, /* completed sync */
177 FIO_Q_QUEUED = 1, /* queued, will complete async */
178 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
181 #define FIO_HDR_MAGIC 0xf00baaef
184 VERIFY_NONE = 0, /* no verification */
185 VERIFY_MD5, /* md5 sum data blocks */
186 VERIFY_CRC32, /* crc32 sum data blocks */
187 VERIFY_NULL, /* pretend to verify */
191 * A header structure associated with each checksummed data block
193 struct verify_header {
194 unsigned int fio_magic;
196 unsigned int verify_type;
198 char md5_digest[MD5_HASH_WORDS * 4];
203 struct group_run_stats {
204 unsigned long long max_run[2], min_run[2];
205 unsigned long long max_bw[2], min_bw[2];
206 unsigned long long io_kb[2];
207 unsigned long long agg[2];
211 * What type of allocation to use for io buffers
214 MEM_MALLOC = 0, /* ordinary malloc */
215 MEM_SHM, /* use shared memory segments */
216 MEM_SHMHUGE, /* use shared memory segments with huge pages */
217 MEM_MMAP, /* use anonynomous mmap */
218 MEM_MMAPHUGE, /* memory mapped huge file */
222 * The type of object we are working on
225 FIO_TYPE_FILE = 1, /* plain file */
226 FIO_TYPE_BD, /* block device */
227 FIO_TYPE_CHAR, /* character device */
228 FIO_TYPE_PIPE, /* pipe */
231 enum fio_ioengine_flags {
232 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
233 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
234 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
235 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
236 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
239 enum fio_file_flags {
240 FIO_FILE_OPEN = 1 << 0, /* file is open */
241 FIO_FILE_UNLINK = 1 << 1, /* unlink on close */
242 FIO_FILE_CLOSING = 1 << 2, /* file being closed */
243 FIO_FILE_EXISTS = 1 << 3, /* file there */
244 FIO_FILE_EXTEND = 1 << 4, /* needs extend */
245 FIO_FILE_NOSORT = 1 << 5, /* don't sort verify blocks */
246 FIO_FILE_DONE = 1 << 6, /* io completed to this file */
247 FIO_SIZE_KNOWN = 1 << 7, /* size has been set */
251 * Each thread_data structure has a number of files associated with it,
252 * this structure holds state information for a single file.
255 enum fio_filetype filetype;
258 * A file may not be a file descriptor, let the io engine decide
261 unsigned long file_data;
266 * filename and possible memory mapping
272 * size of the file, offset into file, and io size from that offset
274 unsigned long long real_file_size;
275 unsigned long long file_offset;
276 unsigned long long io_size;
278 unsigned long long last_pos;
279 unsigned long long last_completed_pos;
282 * block map for random io
284 unsigned long *file_map;
285 unsigned int num_maps;
286 unsigned int last_free_lookup;
289 enum fio_file_flags flags;
293 * How many depth levels to log
295 #define FIO_IO_U_MAP_NR 8
296 #define FIO_IO_U_LAT_NR 12
307 struct io_log *slat_log;
308 struct io_log *clat_log;
309 struct io_log *bw_log;
312 * bandwidth and latency stats
314 struct io_stat clat_stat[2]; /* completion latency */
315 struct io_stat slat_stat[2]; /* submission latency */
316 struct io_stat bw_stat[2]; /* bandwidth stats */
318 unsigned long long stat_io_bytes[2];
319 struct timeval stat_sample_time[2];
322 * fio system usage accounting
324 struct rusage ru_start;
325 struct rusage ru_end;
326 unsigned long usr_time;
327 unsigned long sys_time;
331 * IO depth and latency stats
333 unsigned int io_u_map[FIO_IO_U_MAP_NR];
334 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
335 unsigned long total_io_u[2];
337 unsigned long long io_bytes[2];
338 unsigned long runtime[2];
339 unsigned long total_run_time;
342 struct thread_options {
350 enum td_ddir td_ddir;
351 unsigned int ddir_nr;
352 unsigned int iodepth;
353 unsigned int iodepth_low;
354 unsigned int iodepth_batch;
356 unsigned long long size;
357 unsigned long long file_size_low;
358 unsigned long long file_size_high;
359 unsigned long long start_offset;
362 unsigned int min_bs[2];
363 unsigned int max_bs[2];
365 unsigned int nr_files;
366 unsigned int open_files;
368 unsigned int odirect;
369 unsigned int invalidate_cache;
370 unsigned int create_serialize;
371 unsigned int create_fsync;
372 unsigned int end_fsync;
373 unsigned int sync_io;
375 unsigned int verifysort;
376 unsigned int use_thread;
378 unsigned int do_disk_util;
379 unsigned int override_sync;
380 unsigned int rand_repeatable;
381 unsigned int write_lat_log;
382 unsigned int write_bw_log;
383 unsigned int norandommap;
384 unsigned int bs_unaligned;
385 unsigned int fsync_on_close;
387 unsigned int hugepage_size;
388 unsigned int rw_min_bs;
389 unsigned int thinktime;
390 unsigned int thinktime_spin;
391 unsigned int thinktime_blocks;
392 unsigned int fsync_blocks;
393 unsigned int start_delay;
394 unsigned long timeout;
395 unsigned int overwrite;
396 unsigned int bw_avg_time;
398 unsigned long long zone_size;
399 unsigned long long zone_skip;
400 enum fio_memtype mem_type;
402 unsigned int stonewall;
403 unsigned int new_group;
404 unsigned int numjobs;
405 os_cpu_mask_t cpumask;
407 unsigned int read_iolog;
408 unsigned int rwmixcycle;
409 unsigned int rwmix[2];
411 unsigned int file_service_type;
412 unsigned int group_reporting;
413 unsigned int fadvise_hint;
415 char *read_iolog_file;
416 char *write_iolog_file;
419 * Pre-run and post-run shell
425 unsigned int ratemin;
426 unsigned int ratecycle;
427 unsigned int rate_iops;
428 unsigned int rate_iops_min;
433 * CPU "io" cycle burner
435 unsigned int cpuload;
436 unsigned int cpucycle;
439 #define FIO_VERROR_SIZE 128
442 * This describes a single thread/process executing a fio job.
445 struct thread_options o;
446 char verror[FIO_VERROR_SIZE];
450 struct thread_stat ts;
451 struct fio_file *files;
452 unsigned int files_index;
453 unsigned int nr_open_files;
454 unsigned int nr_done_files;
455 unsigned int nr_normal_files;
457 unsigned int next_file;
458 os_random_state_t next_file_state;
463 size_t orig_buffer_size;
464 volatile int terminate;
465 volatile int runstate;
467 unsigned int last_was_sync;
477 os_random_state_t bsrange_state;
478 os_random_state_t verify_state;
483 * IO engine hooks, contains everything needed to submit an io_u
484 * to any of the available IO engines.
486 struct ioengine_ops *io_ops;
489 * Current IO depth and list of free and busy io_u's.
491 unsigned int cur_depth;
492 struct list_head io_u_freelist;
493 struct list_head io_u_busylist;
494 struct list_head io_u_requeues;
495 unsigned int io_u_queued;
500 unsigned long rate_usec_cycle;
501 long rate_pending_usleep;
502 unsigned long rate_bytes;
503 unsigned long rate_blocks;
504 struct timeval lastrate;
506 unsigned long long total_io_size;
508 unsigned long io_issues[2];
509 unsigned long long io_blocks[2];
510 unsigned long long io_bytes[2];
511 unsigned long long this_io_bytes[2];
512 unsigned long long zone_bytes;
513 struct fio_sem *mutex;
516 * State for random io, a bitmap of blocks done vs not done
518 os_random_state_t random_state;
520 struct timeval start; /* start of this loop */
521 struct timeval epoch; /* time job was started */
522 struct timeval rw_end[2];
523 unsigned int rw_end_set[2];
526 * read/write mixed workload state
528 os_random_state_t rwmix_state;
529 unsigned long long rwmix_bytes;
530 struct timeval rwmix_switch;
531 enum fio_ddir rwmix_ddir;
532 unsigned int ddir_nr;
535 * IO history logs for verification. We use a tree for sorting,
536 * if we are overwriting. Otherwise just use a fifo.
538 struct rb_root io_hist_tree;
539 struct list_head io_hist_list;
544 struct list_head io_log_list;
549 struct timeval timeout_end;
550 struct itimerval timer;
553 * for fileservice, how often to switch to a new file
555 unsigned int file_service_nr;
556 unsigned int file_service_left;
557 struct fio_file *file_service_file;
560 * For generating file sizes
562 os_random_state_t file_size_state;
566 * roundrobin available files, or choose one at random.
569 FIO_FSERVICE_RANDOM = 1,
574 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
575 * the timer on each queue operation.
577 #define IO_U_TIMEOUT_INC 5
578 #define IO_U_TIMEOUT 30
580 #define __td_verror(td, err, msg, func) \
586 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
590 #define td_verror(td, err, func) \
591 __td_verror((td), (err), strerror((err)), (func))
592 #define td_vmsg(td, err, msg, func) \
593 __td_verror((td), (err), (msg), (func))
595 extern int exitall_on_terminate;
596 extern int thread_number;
597 extern int nr_process, nr_thread;
600 extern int terse_output;
603 extern int temp_stall_ts;
604 extern unsigned long long mlock_size;
605 extern unsigned long page_mask, page_size;
607 extern struct thread_data *threads;
609 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
610 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
611 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
612 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
614 #define BLOCKS_PER_MAP (8 * sizeof(long))
615 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
616 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
617 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
619 #define MAX_JOBS (1024)
621 static inline int should_fsync(struct thread_data *td)
623 if (td->last_was_sync)
627 if (td_write(td) || td_rw(td) || td->o.override_sync)
634 * Disk utils as read in /sys/block/<dev>/stat
636 struct disk_util_stat {
639 unsigned long long sectors[2];
642 unsigned time_in_queue;
646 * Per-device disk util management
649 struct list_head list;
656 struct disk_util_stat dus;
657 struct disk_util_stat last_dus;
663 #define DISK_UTIL_MSEC (250)
666 #define min(a, b) ((a) < (b) ? (a) : (b))
669 #define max(a, b) ((a) > (b) ? (a) : (b))
675 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
676 extern void write_iolog_put(struct thread_data *, struct io_u *);
677 extern int __must_check init_iolog(struct thread_data *td);
678 extern void log_io_piece(struct thread_data *, struct io_u *);
679 extern void prune_io_piece_log(struct thread_data *);
680 extern void write_iolog_close(struct thread_data *);
685 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
686 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
687 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
688 extern void show_run_stats(void);
689 extern void init_disk_util(struct thread_data *);
690 extern void update_rusage_stat(struct thread_data *);
691 extern void update_io_ticks(void);
692 extern void disk_util_timer_arm(void);
693 extern void setup_log(struct io_log **);
694 extern void finish_log(struct thread_data *, struct io_log *, const char *);
695 extern void __finish_log(struct io_log *, const char *);
696 extern struct io_log *agg_io_log[2];
697 extern int write_bw_log;
698 extern void add_agg_sample(unsigned long, enum fio_ddir);
703 extern unsigned long utime_since(struct timeval *, struct timeval *);
704 extern unsigned long utime_since_now(struct timeval *);
705 extern unsigned long mtime_since(struct timeval *, struct timeval *);
706 extern unsigned long mtime_since_now(struct timeval *);
707 extern unsigned long time_since_now(struct timeval *);
708 extern unsigned long mtime_since_genesis(void);
709 extern void __usec_sleep(unsigned int);
710 extern void usec_sleep(struct thread_data *, unsigned long);
711 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
712 extern void fill_start_time(struct timeval *);
713 extern void fio_gettime(struct timeval *, void *);
714 extern void set_genesis_time(void);
717 * Init/option functions
719 extern int __must_check parse_options(int, char **);
720 extern int fio_option_parse(struct thread_data *, const char *);
721 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
722 extern void fio_fill_default_options(struct thread_data *);
723 extern int fio_show_option_help(const char *);
724 extern void fio_options_dup_and_init(struct option *);
725 extern void options_mem_dupe(struct thread_data *);
726 extern void options_mem_free(struct thread_data *);
727 #define FIO_GETOPT_JOB 0x89988998
728 #define FIO_NR_OPTIONS 128
731 * File setup/shutdown
733 extern void close_files(struct thread_data *);
734 extern int __must_check setup_files(struct thread_data *);
735 extern int __must_check open_files(struct thread_data *);
736 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
737 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
738 extern void generic_close_file(struct thread_data *, struct fio_file *);
739 extern void add_file(struct thread_data *, const char *);
740 extern void get_file(struct fio_file *);
741 extern void put_file(struct thread_data *, struct fio_file *);
742 extern int add_dir_files(struct thread_data *, const char *);
743 extern int init_random_map(struct thread_data *);
744 extern void dup_files(struct thread_data *, struct thread_data *);
749 extern void print_thread_status(void);
750 extern void print_status_init(int);
755 #ifdef FIO_HAVE_DISK_UTIL
756 extern void show_disk_util(void);
757 extern void disk_util_timer_arm(void);
758 extern void init_disk_util(struct thread_data *);
759 extern void update_io_ticks(void);
761 #define show_disk_util()
762 #define disk_util_timer_arm()
763 #define init_disk_util(td)
764 #define update_io_ticks()
768 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
769 * will never back again. It may cycle between running/verififying/fsyncing.
770 * Once the thread reaches TD_EXITED, it is just waiting for the core to
787 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
788 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
789 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
794 extern int __must_check fio_pin_memory(void);
795 extern void fio_unpin_memory(void);
796 extern int __must_check allocate_io_mem(struct thread_data *);
797 extern void free_io_mem(struct thread_data *);
802 #define queue_full(td) list_empty(&(td)->io_u_freelist)
803 extern struct io_u *__get_io_u(struct thread_data *);
804 extern struct io_u *get_io_u(struct thread_data *);
805 extern void put_io_u(struct thread_data *, struct io_u *);
806 extern void requeue_io_u(struct thread_data *, struct io_u **);
807 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
808 extern long __must_check io_u_queued_complete(struct thread_data *, int);
809 extern void io_u_queued(struct thread_data *, struct io_u *);
810 extern void io_u_log_error(struct thread_data *, struct io_u *);
811 extern void io_u_init_timeout(void);
812 extern void io_u_set_timeout(struct thread_data *);
813 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
816 * io engine entry points
818 extern int __must_check td_io_init(struct thread_data *);
819 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
820 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
821 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
822 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
823 extern int __must_check td_io_commit(struct thread_data *);
824 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
825 extern void td_io_close_file(struct thread_data *, struct fio_file *);
828 * If logging output to a file, stderr should go to both stderr and f_err
830 #define log_err(args...) do { \
831 fprintf(f_err, ##args); \
832 if (f_err != stderr) \
833 fprintf(stderr, ##args); \
836 #define log_info(args...) fprintf(f_out, ##args)
838 FILE *get_f_out(void);
839 FILE *get_f_err(void);
841 struct ioengine_ops {
842 struct list_head list;
846 int (*setup)(struct thread_data *);
847 int (*init)(struct thread_data *);
848 int (*prep)(struct thread_data *, struct io_u *);
849 int (*queue)(struct thread_data *, struct io_u *);
850 int (*commit)(struct thread_data *);
851 int (*getevents)(struct thread_data *, int, int, struct timespec *);
852 struct io_u *(*event)(struct thread_data *, int);
853 int (*cancel)(struct thread_data *, struct io_u *);
854 void (*cleanup)(struct thread_data *);
855 int (*open_file)(struct thread_data *, struct fio_file *);
856 void (*close_file)(struct thread_data *, struct fio_file *);
861 #define FIO_IOOPS_VERSION 7
863 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
864 extern void register_ioengine(struct ioengine_ops *);
865 extern void unregister_ioengine(struct ioengine_ops *);
866 extern void close_ioengine(struct thread_data *);
869 * Mark unused variables passed to ops functions as unused, to silence gcc
871 #define fio_unused __attribute((__unused__))
872 #define fio_init __attribute__((constructor))
873 #define fio_exit __attribute__((destructor))
875 #define for_each_td(td, i) \
876 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
877 #define for_each_file(td, f, i) \
878 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
880 #define fio_assert(td, cond) do { \
883 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
884 (td)->runstate = TD_EXITED; \
885 (td)->error = EFAULT; \
890 static inline void clear_error(struct thread_data *td)
893 td->verror[0] = '\0';