8 #include <sys/resource.h>
18 #include "compiler/compiler.h"
22 #include "arch/arch.h"
38 TD_DDIR_READ = 1 << 0,
39 TD_DDIR_WRITE = 1 << 1,
40 TD_DDIR_RAND = 1 << 2,
41 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
42 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
43 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
44 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
48 * Use for maintaining statistics
51 unsigned long max_val;
52 unsigned long min_val;
53 unsigned long samples;
60 * A single data sample
69 * Dynamically growing data sample log
72 unsigned long nr_samples;
73 unsigned long max_samples;
74 struct io_sample *log;
78 * When logging io actions, this matches a single sent io_u
82 struct rb_node rb_node;
83 struct list_head list;
87 struct fio_file *file;
89 unsigned long long offset;
94 unsigned int file_action;
100 IO_U_F_FLIGHT = 1 << 1,
110 #ifdef FIO_HAVE_LIBAIO
113 #ifdef FIO_HAVE_POSIXAIO
117 struct sg_io_hdr hdr;
119 #ifdef FIO_HAVE_GUASI
123 struct timeval start_time;
124 struct timeval issue_time;
127 * Allocated/set buffer and length
130 unsigned long buflen;
131 unsigned long long offset;
132 unsigned long long endpos;
135 * IO engine state, may be different from above when we get
136 * partial transfers / residual data counts
139 unsigned long xfer_buflen;
147 * io engine private data
156 struct fio_file *file;
158 struct list_head list;
161 * Callback for io completion
163 int (*end_io)(struct thread_data *, struct io_u *);
167 * io_ops->queue() return values
170 FIO_Q_COMPLETED = 0, /* completed sync */
171 FIO_Q_QUEUED = 1, /* queued, will complete async */
172 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
175 #define FIO_HDR_MAGIC 0xf00baaef
178 VERIFY_NONE = 0, /* no verification */
179 VERIFY_MD5, /* md5 sum data blocks */
180 VERIFY_CRC64, /* crc64 sum data blocks */
181 VERIFY_CRC32, /* crc32 sum data blocks */
182 VERIFY_CRC16, /* crc16 sum data blocks */
183 VERIFY_CRC7, /* crc7 sum data blocks */
184 VERIFY_SHA256, /* sha256 sum data blocks */
185 VERIFY_SHA512, /* sha512 sum data blocks */
186 VERIFY_META, /* block_num, timestamp etc. */
187 VERIFY_NULL, /* pretend to verify */
191 * A header structure associated with each checksummed data block. It is
192 * followed by a checksum specific header that contains the verification
195 struct verify_header {
196 unsigned int fio_magic;
198 unsigned int verify_type;
202 uint32_t md5_digest[16];
224 unsigned char thread;
225 unsigned short numberio;
226 unsigned long time_sec;
227 unsigned long time_usec;
230 struct group_run_stats {
231 unsigned long long max_run[2], min_run[2];
232 unsigned long long max_bw[2], min_bw[2];
233 unsigned long long io_kb[2];
234 unsigned long long agg[2];
238 * What type of allocation to use for io buffers
241 MEM_MALLOC = 0, /* ordinary malloc */
242 MEM_SHM, /* use shared memory segments */
243 MEM_SHMHUGE, /* use shared memory segments with huge pages */
244 MEM_MMAP, /* use anonynomous mmap */
245 MEM_MMAPHUGE, /* memory mapped huge file */
249 * The type of object we are working on
252 FIO_TYPE_FILE = 1, /* plain file */
253 FIO_TYPE_BD, /* block device */
254 FIO_TYPE_CHAR, /* character device */
255 FIO_TYPE_PIPE, /* pipe */
258 enum fio_ioengine_flags {
259 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
260 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
261 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
262 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
263 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
264 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
265 FIO_NOIO = 1 << 6, /* thread does only pseudo IO */
268 enum fio_file_flags {
269 FIO_FILE_OPEN = 1 << 0, /* file is open */
270 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
271 FIO_FILE_EXISTS = 1 << 2, /* file there */
272 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
273 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
274 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
275 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
279 * Each thread_data structure has a number of files associated with it,
280 * this structure holds state information for a single file.
283 enum fio_filetype filetype;
286 * A file may not be a file descriptor, let the io engine decide
289 unsigned long file_data;
294 * filename and possible memory mapping
298 unsigned int major, minor;
301 * size of the file, offset into file, and io size from that offset
303 unsigned long long real_file_size;
304 unsigned long long file_offset;
305 unsigned long long io_size;
307 unsigned long long last_pos;
308 unsigned long long last_completed_pos;
311 * block map for random io
313 unsigned long *file_map;
314 unsigned int num_maps;
315 unsigned int last_free_lookup;
318 enum fio_file_flags flags;
322 * How many depth levels to log
324 #define FIO_IO_U_MAP_NR 8
325 #define FIO_IO_U_LAT_U_NR 10
326 #define FIO_IO_U_LAT_M_NR 12
337 struct io_log *slat_log;
338 struct io_log *clat_log;
339 struct io_log *bw_log;
342 * bandwidth and latency stats
344 struct io_stat clat_stat[2]; /* completion latency */
345 struct io_stat slat_stat[2]; /* submission latency */
346 struct io_stat bw_stat[2]; /* bandwidth stats */
348 unsigned long long stat_io_bytes[2];
349 struct timeval stat_sample_time[2];
352 * fio system usage accounting
354 struct rusage ru_start;
355 struct rusage ru_end;
356 unsigned long usr_time;
357 unsigned long sys_time;
359 unsigned long minf, majf;
362 * IO depth and latency stats
364 unsigned int io_u_map[FIO_IO_U_MAP_NR];
365 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
366 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
367 unsigned long total_io_u[2];
368 unsigned long short_io_u[2];
370 unsigned long long io_bytes[2];
371 unsigned long runtime[2];
372 unsigned long total_run_time;
380 struct thread_options {
388 enum td_ddir td_ddir;
389 unsigned int ddir_nr;
390 unsigned int iodepth;
391 unsigned int iodepth_low;
392 unsigned int iodepth_batch;
394 unsigned long long size;
395 unsigned int fill_device;
396 unsigned long long file_size_low;
397 unsigned long long file_size_high;
398 unsigned long long start_offset;
401 unsigned int min_bs[2];
402 unsigned int max_bs[2];
403 struct bssplit *bssplit;
404 unsigned int bssplit_nr;
406 unsigned int nr_files;
407 unsigned int open_files;
409 unsigned int odirect;
410 unsigned int invalidate_cache;
411 unsigned int create_serialize;
412 unsigned int create_fsync;
413 unsigned int end_fsync;
414 unsigned int sync_io;
416 unsigned int do_verify;
417 unsigned int verifysort;
418 unsigned int verify_interval;
419 unsigned int verify_offset;
420 unsigned int verify_pattern;
421 unsigned int verify_pattern_bytes;
422 unsigned int verify_fatal;
423 unsigned int use_thread;
425 unsigned int do_disk_util;
426 unsigned int override_sync;
427 unsigned int rand_repeatable;
428 unsigned int write_lat_log;
429 unsigned int write_bw_log;
430 unsigned int norandommap;
431 unsigned int bs_unaligned;
432 unsigned int fsync_on_close;
434 unsigned int hugepage_size;
435 unsigned int rw_min_bs;
436 unsigned int thinktime;
437 unsigned int thinktime_spin;
438 unsigned int thinktime_blocks;
439 unsigned int fsync_blocks;
440 unsigned int start_delay;
441 unsigned long long timeout;
442 unsigned int overwrite;
443 unsigned int bw_avg_time;
445 unsigned long long zone_size;
446 unsigned long long zone_skip;
447 enum fio_memtype mem_type;
449 unsigned int stonewall;
450 unsigned int new_group;
451 unsigned int numjobs;
452 os_cpu_mask_t cpumask;
453 unsigned int cpumask_set;
455 unsigned int rwmixcycle;
456 unsigned int rwmix[2];
458 unsigned int file_service_type;
459 unsigned int group_reporting;
460 unsigned int fadvise_hint;
461 unsigned int zero_buffers;
462 unsigned int time_based;
464 char *read_iolog_file;
465 char *write_iolog_file;
468 * Pre-run and post-run shell
474 unsigned int ratemin;
475 unsigned int ratecycle;
476 unsigned int rate_iops;
477 unsigned int rate_iops_min;
482 * CPU "io" cycle burner
484 unsigned int cpuload;
485 unsigned int cpucycle;
488 #define FIO_VERROR_SIZE 128
491 * This describes a single thread/process executing a fio job.
494 struct thread_options o;
495 char verror[FIO_VERROR_SIZE];
499 struct thread_stat ts;
500 struct fio_file *files;
501 unsigned int files_index;
502 unsigned int nr_open_files;
503 unsigned int nr_done_files;
504 unsigned int nr_normal_files;
506 unsigned int next_file;
507 os_random_state_t next_file_state;
513 size_t orig_buffer_size;
514 volatile int terminate;
515 volatile int runstate;
517 unsigned int ioprio_set;
518 unsigned int last_was_sync;
528 os_random_state_t bsrange_state;
529 os_random_state_t verify_state;
534 * IO engine hooks, contains everything needed to submit an io_u
535 * to any of the available IO engines.
537 struct ioengine_ops *io_ops;
540 * Current IO depth and list of free and busy io_u's.
542 unsigned int cur_depth;
543 struct list_head io_u_freelist;
544 struct list_head io_u_busylist;
545 struct list_head io_u_requeues;
546 unsigned int io_u_queued;
551 unsigned long rate_usec_cycle;
552 long rate_pending_usleep;
553 unsigned long rate_bytes;
554 unsigned long rate_blocks;
555 struct timeval lastrate;
557 unsigned long long total_io_size;
559 unsigned long io_issues[2];
560 unsigned long long io_blocks[2];
561 unsigned long long io_bytes[2];
562 unsigned long long io_skip_bytes;
563 unsigned long long this_io_bytes[2];
564 unsigned long long zone_bytes;
565 struct fio_sem *mutex;
568 * State for random io, a bitmap of blocks done vs not done
570 os_random_state_t random_state;
572 struct timeval start; /* start of this loop */
573 struct timeval epoch; /* time job was started */
574 struct timeval rw_end[2];
575 struct timeval last_issue;
576 unsigned int rw_end_set[2];
579 * read/write mixed workload state
581 os_random_state_t rwmix_state;
582 unsigned long long rwmix_bytes;
583 struct timeval rwmix_switch;
584 enum fio_ddir rwmix_ddir;
585 unsigned int ddir_nr;
588 * IO history logs for verification. We use a tree for sorting,
589 * if we are overwriting. Otherwise just use a fifo.
591 struct rb_root io_hist_tree;
592 struct list_head io_hist_list;
597 struct list_head io_log_list;
602 struct timeval timeout_end;
603 struct itimerval timer;
606 * for fileservice, how often to switch to a new file
608 unsigned int file_service_nr;
609 unsigned int file_service_left;
610 struct fio_file *file_service_file;
613 * For generating file sizes
615 os_random_state_t file_size_state;
619 * roundrobin available files, or choose one at random.
622 FIO_FSERVICE_RANDOM = 1,
627 * when should interactive ETA output be generated
636 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
637 * the timer on each queue operation.
639 #define IO_U_TIMEOUT_INC 5
640 #define IO_U_TIMEOUT 30
642 #define __td_verror(td, err, msg, func) \
648 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
652 #define td_verror(td, err, func) \
653 __td_verror((td), (err), strerror((err)), (func))
654 #define td_vmsg(td, err, msg, func) \
655 __td_verror((td), (err), (msg), (func))
657 extern int exitall_on_terminate;
658 extern int thread_number;
659 extern int nr_process, nr_thread;
662 extern int terse_output;
665 extern int temp_stall_ts;
666 extern unsigned long long mlock_size;
667 extern unsigned long page_mask, page_size;
668 extern int read_only;
669 extern int eta_print;
671 extern struct thread_data *threads;
673 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
674 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
675 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
676 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
678 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
680 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
683 #define BLOCKS_PER_MAP (8 * sizeof(long))
684 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (unsigned long long) (td)->o.rw_min_bs))
685 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
686 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
688 #define MAX_JOBS (1024)
690 static inline int should_fsync(struct thread_data *td)
692 if (td->last_was_sync)
696 if (td_write(td) || td_rw(td) || td->o.override_sync)
703 * Disk utils as read in /sys/block/<dev>/stat
705 struct disk_util_stat {
708 unsigned long long sectors[2];
711 unsigned time_in_queue;
715 * Per-device disk util management
718 struct list_head list;
725 struct disk_util_stat dus;
726 struct disk_util_stat last_dus;
732 #define DISK_UTIL_MSEC (250)
743 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
744 extern void log_io_u(struct thread_data *, struct io_u *);
745 extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
746 extern int __must_check init_iolog(struct thread_data *td);
747 extern void log_io_piece(struct thread_data *, struct io_u *);
748 extern void prune_io_piece_log(struct thread_data *);
749 extern void write_iolog_close(struct thread_data *);
754 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
755 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
756 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
757 extern void show_run_stats(void);
758 extern void init_disk_util(struct thread_data *);
759 extern void update_rusage_stat(struct thread_data *);
760 extern void update_io_ticks(void);
761 extern void disk_util_timer_arm(void);
762 extern void setup_log(struct io_log **);
763 extern void finish_log(struct thread_data *, struct io_log *, const char *);
764 extern void __finish_log(struct io_log *, const char *);
765 extern struct io_log *agg_io_log[2];
766 extern int write_bw_log;
767 extern void add_agg_sample(unsigned long, enum fio_ddir);
772 extern unsigned long long utime_since(struct timeval *, struct timeval *);
773 extern unsigned long long utime_since_now(struct timeval *);
774 extern unsigned long mtime_since(struct timeval *, struct timeval *);
775 extern unsigned long mtime_since_now(struct timeval *);
776 extern unsigned long time_since_now(struct timeval *);
777 extern unsigned long mtime_since_genesis(void);
778 extern void __usec_sleep(unsigned int);
779 extern void usec_sleep(struct thread_data *, unsigned long);
780 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
781 extern void fill_start_time(struct timeval *);
782 extern void fio_gettime(struct timeval *, void *);
783 extern void set_genesis_time(void);
786 * Init/option functions
788 extern int __must_check parse_options(int, char **);
789 extern int fio_option_parse(struct thread_data *, const char *);
790 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
791 extern void fio_fill_default_options(struct thread_data *);
792 extern int fio_show_option_help(const char *);
793 extern void fio_options_dup_and_init(struct option *);
794 extern void options_mem_dupe(struct thread_data *);
795 extern void options_mem_free(struct thread_data *);
796 #define FIO_GETOPT_JOB 0x89988998
797 #define FIO_NR_OPTIONS 128
800 * File setup/shutdown
802 extern void close_files(struct thread_data *);
803 extern int __must_check setup_files(struct thread_data *);
804 extern int __must_check open_files(struct thread_data *);
805 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
806 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
807 extern void generic_close_file(struct thread_data *, struct fio_file *);
808 extern int add_file(struct thread_data *, const char *);
809 extern void get_file(struct fio_file *);
810 extern void put_file(struct thread_data *, struct fio_file *);
811 extern int add_dir_files(struct thread_data *, const char *);
812 extern int init_random_map(struct thread_data *);
813 extern void dup_files(struct thread_data *, struct thread_data *);
814 extern int get_fileno(struct thread_data *, const char *);
815 extern void free_release_files(struct thread_data *);
820 extern void print_thread_status(void);
821 extern void print_status_init(int);
826 #ifdef FIO_HAVE_DISK_UTIL
827 extern void show_disk_util(void);
828 extern void disk_util_timer_arm(void);
829 extern void init_disk_util(struct thread_data *);
830 extern void update_io_ticks(void);
832 #define show_disk_util()
833 #define disk_util_timer_arm()
834 #define init_disk_util(td)
835 #define update_io_ticks()
839 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
840 * will never back again. It may cycle between running/verififying/fsyncing.
841 * Once the thread reaches TD_EXITED, it is just waiting for the core to
858 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
859 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
860 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
865 extern int __must_check fio_pin_memory(void);
866 extern void fio_unpin_memory(void);
867 extern int __must_check allocate_io_mem(struct thread_data *);
868 extern void free_io_mem(struct thread_data *);
873 #define queue_full(td) list_empty(&(td)->io_u_freelist)
874 extern struct io_u *__get_io_u(struct thread_data *);
875 extern struct io_u *get_io_u(struct thread_data *);
876 extern void put_io_u(struct thread_data *, struct io_u *);
877 extern void requeue_io_u(struct thread_data *, struct io_u **);
878 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
879 extern long __must_check io_u_queued_complete(struct thread_data *, int);
880 extern void io_u_queued(struct thread_data *, struct io_u *);
881 extern void io_u_log_error(struct thread_data *, struct io_u *);
882 extern void io_u_init_timeout(void);
883 extern void io_u_set_timeout(struct thread_data *);
884 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
887 * io engine entry points
889 extern int __must_check td_io_init(struct thread_data *);
890 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
891 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
892 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
893 extern int __must_check td_io_getevents(struct thread_data *, unsigned int, unsigned int, struct timespec *);
894 extern int __must_check td_io_commit(struct thread_data *);
895 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
896 extern void td_io_close_file(struct thread_data *, struct fio_file *);
901 #ifdef FIO_HAVE_BLKTRACE
902 extern int is_blktrace(const char *);
903 extern int load_blktrace(struct thread_data *, const char *);
907 * If logging output to a file, stderr should go to both stderr and f_err
909 #define log_err(args...) do { \
910 fprintf(f_err, ##args); \
911 if (f_err != stderr) \
912 fprintf(stderr, ##args); \
915 #define log_info(args...) fprintf(f_out, ##args)
917 FILE *get_f_out(void);
918 FILE *get_f_err(void);
920 struct ioengine_ops {
921 struct list_head list;
925 int (*setup)(struct thread_data *);
926 int (*init)(struct thread_data *);
927 int (*prep)(struct thread_data *, struct io_u *);
928 int (*queue)(struct thread_data *, struct io_u *);
929 int (*commit)(struct thread_data *);
930 int (*getevents)(struct thread_data *, unsigned int, unsigned int, struct timespec *);
931 struct io_u *(*event)(struct thread_data *, int);
932 int (*cancel)(struct thread_data *, struct io_u *);
933 void (*cleanup)(struct thread_data *);
934 int (*open_file)(struct thread_data *, struct fio_file *);
935 void (*close_file)(struct thread_data *, struct fio_file *);
940 #define FIO_IOOPS_VERSION 8
942 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
943 extern void register_ioengine(struct ioengine_ops *);
944 extern void unregister_ioengine(struct ioengine_ops *);
945 extern void close_ioengine(struct thread_data *);
948 * Mark unused variables passed to ops functions as unused, to silence gcc
950 #define fio_unused __attribute((__unused__))
951 #define fio_init __attribute__((constructor))
952 #define fio_exit __attribute__((destructor))
954 #define for_each_td(td, i) \
955 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
956 #define for_each_file(td, f, i) \
957 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
959 #define fio_assert(td, cond) do { \
962 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
963 (td)->runstate = TD_EXITED; \
964 (td)->error = EFAULT; \
969 static inline void clear_error(struct thread_data *td)
972 td->verror[0] = '\0';
990 extern struct debug_level debug_levels[];
992 extern unsigned long fio_debug;
994 #define dprint(type, str, args...) \
996 assert(type < FD_DEBUG_MAX); \
997 if ((((1 << type)) & fio_debug) == 0) \
999 log_info("%-8s ", debug_levels[(type)].name); \
1000 log_info(str, ##args); \
1003 static inline void dprint_io_u(struct io_u *io_u, const char *p)
1005 struct fio_file *f = io_u->file;
1007 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
1008 io_u->offset, io_u->buflen, io_u->ddir);
1009 if (fio_debug & (1 << FD_IO)) {
1011 log_info("/%s", f->file_name);
1017 #define dprint(type, str, args...)
1018 #define dprint_io_u(io_u, p)