8 #include <sys/resource.h>
16 #include "compiler/compiler.h"
22 #include "arch/arch.h"
26 #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;
88 struct fio_file *file;
89 unsigned long long offset;
95 #ifdef FIO_HAVE_SYSLET
97 struct syslet_uatom atom; /* the atom to submit */
98 struct syslet_uatom *head; /* head of the sequence */
99 long ret; /* syscall return value */
104 IO_U_F_FREE = 1 << 0,
105 IO_U_F_FLIGHT = 1 << 1,
115 #ifdef FIO_HAVE_LIBAIO
118 #ifdef FIO_HAVE_POSIXAIO
122 struct sg_io_hdr hdr;
124 #ifdef FIO_HAVE_SYSLET
125 struct syslet_req req;
127 #ifdef FIO_HAVE_GUASI
131 struct timeval start_time;
132 struct timeval issue_time;
135 * Allocated/set buffer and length
138 unsigned long buflen;
139 unsigned long long offset;
140 unsigned long long endpos;
143 * IO engine state, may be different from above when we get
144 * partial transfers / residual data counts
147 unsigned long xfer_buflen;
155 * io engine private data
164 struct fio_file *file;
166 struct list_head list;
169 * Callback for io completion
171 int (*end_io)(struct thread_data *, struct io_u *);
175 * io_ops->queue() return values
178 FIO_Q_COMPLETED = 0, /* completed sync */
179 FIO_Q_QUEUED = 1, /* queued, will complete async */
180 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
183 #define FIO_HDR_MAGIC 0xf00baaef
186 VERIFY_NONE = 0, /* no verification */
187 VERIFY_MD5, /* md5 sum data blocks */
188 VERIFY_CRC32, /* crc32 sum data blocks */
189 VERIFY_NULL, /* pretend to verify */
193 * A header structure associated with each checksummed data block
195 struct verify_header {
196 unsigned int fio_magic;
198 unsigned int verify_type;
200 char md5_digest[MD5_HASH_WORDS * 4];
205 struct group_run_stats {
206 unsigned long long max_run[2], min_run[2];
207 unsigned long long max_bw[2], min_bw[2];
208 unsigned long long io_kb[2];
209 unsigned long long agg[2];
213 * What type of allocation to use for io buffers
216 MEM_MALLOC = 0, /* ordinary malloc */
217 MEM_SHM, /* use shared memory segments */
218 MEM_SHMHUGE, /* use shared memory segments with huge pages */
219 MEM_MMAP, /* use anonynomous mmap */
220 MEM_MMAPHUGE, /* memory mapped huge file */
224 * The type of object we are working on
227 FIO_TYPE_FILE = 1, /* plain file */
228 FIO_TYPE_BD, /* block device */
229 FIO_TYPE_CHAR, /* character device */
230 FIO_TYPE_PIPE, /* pipe */
233 enum fio_ioengine_flags {
234 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
235 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
236 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
237 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
238 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
241 enum fio_file_flags {
242 FIO_FILE_OPEN = 1 << 0, /* file is open */
243 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
244 FIO_FILE_EXISTS = 1 << 2, /* file there */
245 FIO_FILE_EXTEND = 1 << 3, /* needs extend */
246 FIO_FILE_NOSORT = 1 << 4, /* don't sort verify blocks */
247 FIO_FILE_DONE = 1 << 5, /* io completed to this file */
248 FIO_SIZE_KNOWN = 1 << 6, /* size has been set */
252 * Each thread_data structure has a number of files associated with it,
253 * this structure holds state information for a single file.
256 enum fio_filetype filetype;
259 * A file may not be a file descriptor, let the io engine decide
262 unsigned long file_data;
267 * filename and possible memory mapping
271 unsigned int major, minor;
274 * size of the file, offset into file, and io size from that offset
276 unsigned long long real_file_size;
277 unsigned long long file_offset;
278 unsigned long long io_size;
280 unsigned long long last_pos;
281 unsigned long long last_completed_pos;
284 * block map for random io
286 unsigned long *file_map;
287 unsigned int num_maps;
288 unsigned int last_free_lookup;
291 enum fio_file_flags flags;
295 * How many depth levels to log
297 #define FIO_IO_U_MAP_NR 8
298 #define FIO_IO_U_LAT_NR 12
309 struct io_log *slat_log;
310 struct io_log *clat_log;
311 struct io_log *bw_log;
314 * bandwidth and latency stats
316 struct io_stat clat_stat[2]; /* completion latency */
317 struct io_stat slat_stat[2]; /* submission latency */
318 struct io_stat bw_stat[2]; /* bandwidth stats */
320 unsigned long long stat_io_bytes[2];
321 struct timeval stat_sample_time[2];
324 * fio system usage accounting
326 struct rusage ru_start;
327 struct rusage ru_end;
328 unsigned long usr_time;
329 unsigned long sys_time;
333 * IO depth and latency stats
335 unsigned int io_u_map[FIO_IO_U_MAP_NR];
336 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
337 unsigned long total_io_u[2];
338 unsigned long short_io_u[2];
340 unsigned long long io_bytes[2];
341 unsigned long runtime[2];
342 unsigned long total_run_time;
345 struct thread_options {
353 enum td_ddir td_ddir;
354 unsigned int ddir_nr;
355 unsigned int iodepth;
356 unsigned int iodepth_low;
357 unsigned int iodepth_batch;
359 unsigned long long size;
360 unsigned long long file_size_low;
361 unsigned long long file_size_high;
362 unsigned long long start_offset;
365 unsigned int min_bs[2];
366 unsigned int max_bs[2];
368 unsigned int nr_files;
369 unsigned int open_files;
371 unsigned int odirect;
372 unsigned int invalidate_cache;
373 unsigned int create_serialize;
374 unsigned int create_fsync;
375 unsigned int end_fsync;
376 unsigned int sync_io;
378 unsigned int verifysort;
379 unsigned int use_thread;
381 unsigned int do_disk_util;
382 unsigned int override_sync;
383 unsigned int rand_repeatable;
384 unsigned int write_lat_log;
385 unsigned int write_bw_log;
386 unsigned int norandommap;
387 unsigned int bs_unaligned;
388 unsigned int fsync_on_close;
390 unsigned int hugepage_size;
391 unsigned int rw_min_bs;
392 unsigned int thinktime;
393 unsigned int thinktime_spin;
394 unsigned int thinktime_blocks;
395 unsigned int fsync_blocks;
396 unsigned int start_delay;
397 unsigned long long timeout;
398 unsigned int overwrite;
399 unsigned int bw_avg_time;
401 unsigned long long zone_size;
402 unsigned long long zone_skip;
403 enum fio_memtype mem_type;
405 unsigned int stonewall;
406 unsigned int new_group;
407 unsigned int numjobs;
408 os_cpu_mask_t cpumask;
409 unsigned int cpumask_set;
411 unsigned int rwmixcycle;
412 unsigned int rwmix[2];
414 unsigned int file_service_type;
415 unsigned int group_reporting;
416 unsigned int fadvise_hint;
417 unsigned int zero_buffers;
418 unsigned int time_based;
420 char *read_iolog_file;
421 char *write_iolog_file;
424 * Pre-run and post-run shell
430 unsigned int ratemin;
431 unsigned int ratecycle;
432 unsigned int rate_iops;
433 unsigned int rate_iops_min;
438 * CPU "io" cycle burner
440 unsigned int cpuload;
441 unsigned int cpucycle;
444 #define FIO_VERROR_SIZE 128
447 * This describes a single thread/process executing a fio job.
450 struct thread_options o;
451 char verror[FIO_VERROR_SIZE];
455 struct thread_stat ts;
456 struct fio_file *files;
457 unsigned int files_index;
458 unsigned int nr_open_files;
459 unsigned int nr_done_files;
460 unsigned int nr_normal_files;
462 unsigned int next_file;
463 os_random_state_t next_file_state;
468 size_t orig_buffer_size;
469 volatile int terminate;
470 volatile int runstate;
472 unsigned int last_was_sync;
482 os_random_state_t bsrange_state;
483 os_random_state_t verify_state;
488 * IO engine hooks, contains everything needed to submit an io_u
489 * to any of the available IO engines.
491 struct ioengine_ops *io_ops;
494 * Current IO depth and list of free and busy io_u's.
496 unsigned int cur_depth;
497 struct list_head io_u_freelist;
498 struct list_head io_u_busylist;
499 struct list_head io_u_requeues;
500 unsigned int io_u_queued;
505 unsigned long rate_usec_cycle;
506 long rate_pending_usleep;
507 unsigned long rate_bytes;
508 unsigned long rate_blocks;
509 struct timeval lastrate;
511 unsigned long long total_io_size;
513 unsigned long io_issues[2];
514 unsigned long long io_blocks[2];
515 unsigned long long io_bytes[2];
516 unsigned long long this_io_bytes[2];
517 unsigned long long zone_bytes;
518 struct fio_sem *mutex;
521 * State for random io, a bitmap of blocks done vs not done
523 os_random_state_t random_state;
525 struct timeval start; /* start of this loop */
526 struct timeval epoch; /* time job was started */
527 struct timeval rw_end[2];
528 struct timeval last_issue;
529 unsigned int rw_end_set[2];
532 * read/write mixed workload state
534 os_random_state_t rwmix_state;
535 unsigned long long rwmix_bytes;
536 struct timeval rwmix_switch;
537 enum fio_ddir rwmix_ddir;
538 unsigned int ddir_nr;
541 * IO history logs for verification. We use a tree for sorting,
542 * if we are overwriting. Otherwise just use a fifo.
544 struct rb_root io_hist_tree;
545 struct list_head io_hist_list;
550 struct list_head io_log_list;
555 struct timeval timeout_end;
556 struct itimerval timer;
559 * for fileservice, how often to switch to a new file
561 unsigned int file_service_nr;
562 unsigned int file_service_left;
563 struct fio_file *file_service_file;
566 * For generating file sizes
568 os_random_state_t file_size_state;
572 * roundrobin available files, or choose one at random.
575 FIO_FSERVICE_RANDOM = 1,
580 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
581 * the timer on each queue operation.
583 #define IO_U_TIMEOUT_INC 5
584 #define IO_U_TIMEOUT 30
586 #define __td_verror(td, err, msg, func) \
592 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
596 #define td_verror(td, err, func) \
597 __td_verror((td), (err), strerror((err)), (func))
598 #define td_vmsg(td, err, msg, func) \
599 __td_verror((td), (err), (msg), (func))
601 extern int exitall_on_terminate;
602 extern int thread_number;
603 extern int nr_process, nr_thread;
606 extern int terse_output;
609 extern int temp_stall_ts;
610 extern unsigned long long mlock_size;
611 extern unsigned long page_mask, page_size;
613 extern struct thread_data *threads;
615 #define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
616 #define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
617 #define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
618 #define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
620 #define BLOCKS_PER_MAP (8 * sizeof(long))
621 #define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->o.rw_min_bs))
622 #define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
623 #define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
625 #define MAX_JOBS (1024)
627 static inline int should_fsync(struct thread_data *td)
629 if (td->last_was_sync)
633 if (td_write(td) || td_rw(td) || td->o.override_sync)
640 * Disk utils as read in /sys/block/<dev>/stat
642 struct disk_util_stat {
645 unsigned long long sectors[2];
648 unsigned time_in_queue;
652 * Per-device disk util management
655 struct list_head list;
662 struct disk_util_stat dus;
663 struct disk_util_stat last_dus;
669 #define DISK_UTIL_MSEC (250)
674 extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
675 extern void write_iolog_put(struct thread_data *, struct io_u *);
676 extern int __must_check init_iolog(struct thread_data *td);
677 extern void log_io_piece(struct thread_data *, struct io_u *);
678 extern void prune_io_piece_log(struct thread_data *);
679 extern void write_iolog_close(struct thread_data *);
684 extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
685 extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
686 extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
687 extern void show_run_stats(void);
688 extern void init_disk_util(struct thread_data *);
689 extern void update_rusage_stat(struct thread_data *);
690 extern void update_io_ticks(void);
691 extern void disk_util_timer_arm(void);
692 extern void setup_log(struct io_log **);
693 extern void finish_log(struct thread_data *, struct io_log *, const char *);
694 extern void __finish_log(struct io_log *, const char *);
695 extern struct io_log *agg_io_log[2];
696 extern int write_bw_log;
697 extern void add_agg_sample(unsigned long, enum fio_ddir);
702 extern unsigned long utime_since(struct timeval *, struct timeval *);
703 extern unsigned long utime_since_now(struct timeval *);
704 extern unsigned long mtime_since(struct timeval *, struct timeval *);
705 extern unsigned long mtime_since_now(struct timeval *);
706 extern unsigned long time_since_now(struct timeval *);
707 extern unsigned long mtime_since_genesis(void);
708 extern void __usec_sleep(unsigned int);
709 extern void usec_sleep(struct thread_data *, unsigned long);
710 extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
711 extern void fill_start_time(struct timeval *);
712 extern void fio_gettime(struct timeval *, void *);
713 extern void set_genesis_time(void);
716 * Init/option functions
718 extern int __must_check parse_options(int, char **);
719 extern int fio_option_parse(struct thread_data *, const char *);
720 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
721 extern void fio_fill_default_options(struct thread_data *);
722 extern int fio_show_option_help(const char *);
723 extern void fio_options_dup_and_init(struct option *);
724 extern void options_mem_dupe(struct thread_data *);
725 extern void options_mem_free(struct thread_data *);
726 #define FIO_GETOPT_JOB 0x89988998
727 #define FIO_NR_OPTIONS 128
730 * File setup/shutdown
732 extern void close_files(struct thread_data *);
733 extern int __must_check setup_files(struct thread_data *);
734 extern int __must_check open_files(struct thread_data *);
735 extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
736 extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
737 extern void generic_close_file(struct thread_data *, struct fio_file *);
738 extern void add_file(struct thread_data *, const char *);
739 extern void get_file(struct fio_file *);
740 extern void put_file(struct thread_data *, struct fio_file *);
741 extern int add_dir_files(struct thread_data *, const char *);
742 extern int init_random_map(struct thread_data *);
743 extern void dup_files(struct thread_data *, struct thread_data *);
748 extern void print_thread_status(void);
749 extern void print_status_init(int);
754 #ifdef FIO_HAVE_DISK_UTIL
755 extern void show_disk_util(void);
756 extern void disk_util_timer_arm(void);
757 extern void init_disk_util(struct thread_data *);
758 extern void update_io_ticks(void);
760 #define show_disk_util()
761 #define disk_util_timer_arm()
762 #define init_disk_util(td)
763 #define update_io_ticks()
767 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
768 * will never back again. It may cycle between running/verififying/fsyncing.
769 * Once the thread reaches TD_EXITED, it is just waiting for the core to
786 extern void populate_verify_io_u(struct thread_data *, struct io_u *);
787 extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
788 extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
793 extern int __must_check fio_pin_memory(void);
794 extern void fio_unpin_memory(void);
795 extern int __must_check allocate_io_mem(struct thread_data *);
796 extern void free_io_mem(struct thread_data *);
801 #define queue_full(td) list_empty(&(td)->io_u_freelist)
802 extern struct io_u *__get_io_u(struct thread_data *);
803 extern struct io_u *get_io_u(struct thread_data *);
804 extern void put_io_u(struct thread_data *, struct io_u *);
805 extern void requeue_io_u(struct thread_data *, struct io_u **);
806 extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
807 extern long __must_check io_u_queued_complete(struct thread_data *, int);
808 extern void io_u_queued(struct thread_data *, struct io_u *);
809 extern void io_u_log_error(struct thread_data *, struct io_u *);
810 extern void io_u_init_timeout(void);
811 extern void io_u_set_timeout(struct thread_data *);
812 extern void io_u_mark_depth(struct thread_data *, struct io_u *);
815 * io engine entry points
817 extern int __must_check td_io_init(struct thread_data *);
818 extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
819 extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
820 extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
821 extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
822 extern int __must_check td_io_commit(struct thread_data *);
823 extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
824 extern void td_io_close_file(struct thread_data *, struct fio_file *);
829 #ifdef FIO_HAVE_BLKTRACE
830 extern int is_blktrace(const char *);
831 extern int load_blktrace(struct thread_data *, const char *);
835 * If logging output to a file, stderr should go to both stderr and f_err
837 #define log_err(args...) do { \
838 fprintf(f_err, ##args); \
839 if (f_err != stderr) \
840 fprintf(stderr, ##args); \
843 #define log_info(args...) fprintf(f_out, ##args)
845 FILE *get_f_out(void);
846 FILE *get_f_err(void);
848 struct ioengine_ops {
849 struct list_head list;
853 int (*setup)(struct thread_data *);
854 int (*init)(struct thread_data *);
855 int (*prep)(struct thread_data *, struct io_u *);
856 int (*queue)(struct thread_data *, struct io_u *);
857 int (*commit)(struct thread_data *);
858 int (*getevents)(struct thread_data *, int, int, struct timespec *);
859 struct io_u *(*event)(struct thread_data *, int);
860 int (*cancel)(struct thread_data *, struct io_u *);
861 void (*cleanup)(struct thread_data *);
862 int (*open_file)(struct thread_data *, struct fio_file *);
863 void (*close_file)(struct thread_data *, struct fio_file *);
868 #define FIO_IOOPS_VERSION 7
870 extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
871 extern void register_ioengine(struct ioengine_ops *);
872 extern void unregister_ioengine(struct ioengine_ops *);
873 extern void close_ioengine(struct thread_data *);
876 * Mark unused variables passed to ops functions as unused, to silence gcc
878 #define fio_unused __attribute((__unused__))
879 #define fio_init __attribute__((constructor))
880 #define fio_exit __attribute__((destructor))
882 #define for_each_td(td, i) \
883 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
884 #define for_each_file(td, f, i) \
885 for ((i) = 0, (f) = &(td)->files[0]; (i) < (td)->o.nr_files; (i)++, (f)++)
887 #define fio_assert(td, cond) do { \
890 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
891 (td)->runstate = TD_EXITED; \
892 (td)->error = EFAULT; \
897 static inline void clear_error(struct thread_data *td)
900 td->verror[0] = '\0';