projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
One step closer to group reporting
[fio.git] / fio.h
... / ...
CommitLineData
1#ifndef FIO_H
2#define FIO_H
3
4#include <sched.h>
5#include <limits.h>
6#include <pthread.h>
7#include <sys/time.h>
8#include <sys/resource.h>
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
12#include <unistd.h>
13#include <string.h>
14
15#include "list.h"
16#include "md5.h"
17#include "crc32.h"
18#include "arch.h"
19#include "os.h"
20
21#ifdef FIO_HAVE_SYSLET
22#include "syslet.h"
23#endif
24
25enum fio_ddir {
26 DDIR_READ = 0,
27 DDIR_WRITE,
28 DDIR_SYNC,
29};
30
31enum td_ddir {
32 TD_DDIR_READ = 1 << 0,
33 TD_DDIR_WRITE = 1 << 1,
34 TD_DDIR_RAND = 1 << 2,
35 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
36 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
37 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
38 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
39};
40
41/*
42 * Use for maintaining statistics
43 */
44struct io_stat {
45 unsigned long max_val;
46 unsigned long min_val;
47 unsigned long samples;
48
49 double mean;
50 double S;
51};
52
53/*
54 * A single data sample
55 */
56struct io_sample {
57 unsigned long time;
58 unsigned long val;
59 enum fio_ddir ddir;
60};
61
62/*
63 * Dynamically growing data sample log
64 */
65struct io_log {
66 unsigned long nr_samples;
67 unsigned long max_samples;
68 struct io_sample *log;
69};
70
71/*
72 * When logging io actions, this matches a single sent io_u
73 */
74struct io_piece {
75 struct list_head list;
76 struct fio_file *file;
77 unsigned long long offset;
78 unsigned long len;
79 enum fio_ddir ddir;
80};
81
82#ifdef FIO_HAVE_SYSLET
83struct syslet_req {
84 struct syslet_uatom atom; /* the atom to submit */
85 struct syslet_uatom *head; /* head of the sequence */
86 long ret; /* syscall return value */
87};
88#endif
89
90enum {
91 IO_U_F_FREE = 1 << 0,
92 IO_U_F_FLIGHT = 1 << 1,
93};
94
95/*
96 * The io unit
97 */
98struct io_u {
99 union {
100#ifdef FIO_HAVE_LIBAIO
101 struct iocb iocb;
102#endif
103#ifdef FIO_HAVE_POSIXAIO
104 struct aiocb aiocb;
105#endif
106#ifdef FIO_HAVE_SGIO
107 struct sg_io_hdr hdr;
108#endif
109#ifdef FIO_HAVE_SYSLET
110 struct syslet_req req;
111#endif
112 };
113 struct timeval start_time;
114 struct timeval issue_time;
115
116 /*
117 * Allocated/set buffer and length
118 */
119 void *buf;
120 unsigned long buflen;
121 unsigned long long offset;
122
123 /*
124 * IO engine state, may be different from above when we get
125 * partial transfers / residual data counts
126 */
127 void *xfer_buf;
128 unsigned long xfer_buflen;
129
130 unsigned int resid;
131 unsigned int error;
132
133 enum fio_ddir ddir;
134
135 /*
136 * io engine private data
137 */
138 union {
139 unsigned int index;
140 unsigned int seen;
141 };
142
143 unsigned int flags;
144
145 struct fio_file *file;
146
147 struct list_head list;
148
149 /*
150 * Callback for io completion
151 */
152 int (*end_io)(struct io_u *);
153};
154
155/*
156 * io_ops->queue() return values
157 */
158enum {
159 FIO_Q_COMPLETED = 0, /* completed sync */
160 FIO_Q_QUEUED = 1, /* queued, will complete async */
161 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
162};
163
164#define FIO_HDR_MAGIC 0xf00baaef
165
166enum {
167 VERIFY_NONE = 0, /* no verification */
168 VERIFY_MD5, /* md5 sum data blocks */
169 VERIFY_CRC32, /* crc32 sum data blocks */
170};
171
172/*
173 * A header structure associated with each checksummed data block
174 */
175struct verify_header {
176 unsigned int fio_magic;
177 unsigned int len;
178 unsigned int verify_type;
179 union {
180 char md5_digest[MD5_HASH_WORDS * 4];
181 unsigned long crc32;
182 };
183};
184
185struct group_run_stats {
186 unsigned long long max_run[2], min_run[2];
187 unsigned long long max_bw[2], min_bw[2];
188 unsigned long long io_kb[2];
189 unsigned long long agg[2];
190};
191
192/*
193 * What type of allocation to use for io buffers
194 */
195enum fio_memtype {
196 MEM_MALLOC = 0, /* ordinary malloc */
197 MEM_SHM, /* use shared memory segments */
198 MEM_SHMHUGE, /* use shared memory segments with huge pages */
199 MEM_MMAP, /* use anonynomous mmap */
200 MEM_MMAPHUGE, /* memory mapped huge file */
201};
202
203/*
204 * The type of object we are working on
205 */
206enum fio_filetype {
207 FIO_TYPE_FILE = 1, /* plain file */
208 FIO_TYPE_BD, /* block device */
209 FIO_TYPE_CHAR, /* character device */
210};
211
212enum fio_ioengine_flags {
213 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
214 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
215 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
216 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
217 FIO_DISKLESSIO = 1 << 4, /* no disk involved */
218 FIO_SELFOPEN = 1 << 5, /* opens its own devices */
219};
220
221/*
222 * Each thread_data structure has a number of files associated with it,
223 * this structure holds state information for a single file.
224 */
225struct fio_file {
226 /*
227 * A file may not be a file descriptor, let the io engine decide
228 */
229 union {
230 unsigned long file_data;
231 int fd;
232 };
233 char *file_name;
234 void *mmap;
235 unsigned long long file_size;
236 unsigned long long real_file_size;
237 unsigned long long file_offset;
238 unsigned long long last_pos;
239 unsigned long long last_completed_pos;
240
241 /*
242 * block map for random io
243 */
244 unsigned long *file_map;
245 unsigned int num_maps;
246 unsigned int last_free_lookup;
247
248 unsigned int unlink;
249};
250
251/*
252 * How many depth levels to log
253 */
254#define FIO_IO_U_MAP_NR 8
255#define FIO_IO_U_LAT_NR 12
256
257struct thread_stat {
258 struct io_log *slat_log;
259 struct io_log *clat_log;
260 struct io_log *bw_log;
261
262 /*
263 * bandwidth and latency stats
264 */
265 struct io_stat clat_stat[2]; /* completion latency */
266 struct io_stat slat_stat[2]; /* submission latency */
267 struct io_stat bw_stat[2]; /* bandwidth stats */
268
269 unsigned long long stat_io_bytes[2];
270 struct timeval stat_sample_time[2];
271
272 /*
273 * fio system usage accounting
274 */
275 struct rusage ru_start;
276 struct rusage ru_end;
277 unsigned long usr_time;
278 unsigned long sys_time;
279 unsigned long ctx;
280
281 /*
282 * IO depth and latency stats
283 */
284 unsigned int io_u_map[FIO_IO_U_MAP_NR];
285 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
286 unsigned long total_io_u;
287};
288
289/*
290 * This describes a single thread/process executing a fio job.
291 */
292struct thread_data {
293 int pad;
294 char *description;
295 char *name;
296 char *directory;
297 char *filename;
298 char verror[128];
299 pthread_t thread;
300 int thread_number;
301 int groupid;
302 struct thread_stat *ts;
303 struct thread_stat __ts;
304 enum fio_filetype filetype;
305 struct fio_file *files;
306 unsigned int nr_files;
307 unsigned int nr_open_files;
308 unsigned int nr_uniq_files;
309 union {
310 unsigned int next_file;
311 os_random_state_t next_file_state;
312 };
313 int error;
314 pid_t pid;
315 char *orig_buffer;
316 size_t orig_buffer_size;
317 volatile int terminate;
318 volatile int runstate;
319 enum td_ddir td_ddir;
320 unsigned int ioprio;
321 unsigned int last_was_sync;
322
323 unsigned int odirect;
324 unsigned int invalidate_cache;
325 unsigned int create_serialize;
326 unsigned int create_fsync;
327 unsigned int end_fsync;
328 unsigned int sync_io;
329 unsigned int verify;
330 unsigned int use_thread;
331 unsigned int unlink;
332 unsigned int do_disk_util;
333 unsigned int override_sync;
334 unsigned int rand_repeatable;
335 unsigned int write_lat_log;
336 unsigned int write_bw_log;
337 unsigned int norandommap;
338 unsigned int bs_unaligned;
339
340 unsigned int bs[2];
341 unsigned int min_bs[2];
342 unsigned int max_bs[2];
343 unsigned int hugepage_size;
344 unsigned int rw_min_bs;
345 unsigned int thinktime;
346 unsigned int thinktime_spin;
347 unsigned int thinktime_blocks;
348 unsigned int fsync_blocks;
349 unsigned int start_delay;
350 unsigned long timeout;
351 unsigned int overwrite;
352 unsigned int bw_avg_time;
353 unsigned int loops;
354 unsigned long long zone_size;
355 unsigned long long zone_skip;
356 enum fio_memtype mem_type;
357 char *mmapfile;
358 int mmapfd;
359 unsigned int stonewall;
360 unsigned int numjobs;
361 unsigned int iodepth;
362 unsigned int iodepth_low;
363 unsigned int iodepth_batch;
364 os_cpu_mask_t cpumask;
365 unsigned int iolog;
366 unsigned int read_iolog;
367 unsigned int rwmixcycle;
368 unsigned int rwmixread;
369 unsigned int rwmixwrite;
370 unsigned int nice;
371 unsigned int file_service_type;
372 unsigned int group_reporting;
373
374 char *read_iolog_file;
375 char *write_iolog_file;
376 void *iolog_buf;
377 FILE *iolog_f;
378
379 char *sysfs_root;
380 char *ioscheduler;
381
382 os_random_state_t bsrange_state;
383 os_random_state_t verify_state;
384
385 int shm_id;
386
387 /*
388 * IO engine hooks, contains everything needed to submit an io_u
389 * to any of the available IO engines.
390 */
391 struct ioengine_ops *io_ops;
392
393 /*
394 * Current IO depth and list of free and busy io_u's.
395 */
396 unsigned int cur_depth;
397 struct list_head io_u_freelist;
398 struct list_head io_u_busylist;
399 struct list_head io_u_requeues;
400 unsigned int io_u_queued;
401
402 /*
403 * Rate state
404 */
405 unsigned int rate;
406 unsigned int ratemin;
407 unsigned int ratecycle;
408 unsigned long rate_usec_cycle;
409 long rate_pending_usleep;
410 unsigned long rate_bytes;
411 struct timeval lastrate;
412
413 unsigned long runtime[2]; /* msec */
414 unsigned long long io_size;
415 unsigned long long total_file_size;
416 unsigned long long start_offset;
417 unsigned long long total_io_size;
418
419 unsigned long io_issues[2];
420 unsigned long long io_blocks[2];
421 unsigned long long io_bytes[2];
422 unsigned long long this_io_bytes[2];
423 unsigned long long zone_bytes;
424 volatile int mutex;
425
426 /*
427 * State for random io, a bitmap of blocks done vs not done
428 */
429 os_random_state_t random_state;
430
431 /*
432 * CPU "io" cycle burner
433 */
434 unsigned int cpuload;
435 unsigned int cpucycle;
436
437 struct timeval start; /* start of this loop */
438 struct timeval epoch; /* time job was started */
439 struct timeval end_time;/* time job ended */
440
441 /*
442 * read/write mixed workload state
443 */
444 os_random_state_t rwmix_state;
445 struct timeval rwmix_switch;
446 enum fio_ddir rwmix_ddir;
447
448 /*
449 * Pre-run and post-run shell
450 */
451 char *exec_prerun;
452 char *exec_postrun;
453
454 /*
455 * IO historic logs
456 */
457 struct list_head io_hist_list;
458 struct list_head io_log_list;
459
460 /*
461 * timeout handling
462 */
463 struct timeval timeout_end;
464 struct itimerval timer;
465};
466
467/*
468 * roundrobin available files, or choose one at random.
469 */
470enum {
471 FIO_FSERVICE_RANDOM = 1,
472 FIO_FSERVICE_RR = 2,
473};
474
475/*
476 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
477 * the timer on each queue operation.
478 */
479#define IO_U_TIMEOUT_INC 5
480#define IO_U_TIMEOUT 30
481
482#define __td_verror(td, err, msg, func) \
483 do { \
484 if ((td)->error) \
485 break; \
486 int e = (err); \
487 (td)->error = e; \
488 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
489 } while (0)
490
491
492#define td_verror(td, err, func) \
493 __td_verror((td), (err), strerror((err)), (func))
494#define td_vmsg(td, err, msg, func) \
495 __td_verror((td), (err), (msg), (func))
496
497extern int exitall_on_terminate;
498extern int thread_number;
499extern int shm_id;
500extern int groupid;
501extern int terse_output;
502extern FILE *f_out;
503extern FILE *f_err;
504extern int temp_stall_ts;
505extern unsigned long long mlock_size;
506
507extern struct thread_data *threads;
508
509#define td_read(td) ((td)->td_ddir & TD_DDIR_READ)
510#define td_write(td) ((td)->td_ddir & TD_DDIR_WRITE)
511#define td_rw(td) (((td)->td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
512#define td_random(td) ((td)->td_ddir & TD_DDIR_RAND)
513
514#define BLOCKS_PER_MAP (8 * sizeof(long))
515#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
516#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
517#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
518
519#define MAX_JOBS (1024)
520
521static inline int should_fsync(struct thread_data *td)
522{
523 if (td->last_was_sync)
524 return 0;
525 if (td->odirect)
526 return 0;
527 if (td_write(td) || td_rw(td) || td->override_sync)
528 return 1;
529
530 return 0;
531}
532
533/*
534 * Disk utils as read in /sys/block/<dev>/stat
535 */
536struct disk_util_stat {
537 unsigned ios[2];
538 unsigned merges[2];
539 unsigned long long sectors[2];
540 unsigned ticks[2];
541 unsigned io_ticks;
542 unsigned time_in_queue;
543};
544
545/*
546 * Per-device disk util management
547 */
548struct disk_util {
549 struct list_head list;
550
551 char *name;
552 char path[256];
553 dev_t dev;
554
555 struct disk_util_stat dus;
556 struct disk_util_stat last_dus;
557
558 unsigned long msec;
559 struct timeval time;
560};
561
562#define DISK_UTIL_MSEC (250)
563
564#ifndef min
565#define min(a, b) ((a) < (b) ? (a) : (b))
566#endif
567#ifndef max
568#define max(a, b) ((a) > (b) ? (a) : (b))
569#endif
570
571/*
572 * Log exports
573 */
574extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
575extern void write_iolog_put(struct thread_data *, struct io_u *);
576extern int __must_check init_iolog(struct thread_data *td);
577extern void log_io_piece(struct thread_data *, struct io_u *);
578extern void prune_io_piece_log(struct thread_data *);
579extern void write_iolog_close(struct thread_data *);
580
581/*
582 * Logging
583 */
584extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
585extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
586extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
587extern void show_run_stats(void);
588extern void init_disk_util(struct thread_data *);
589extern void update_rusage_stat(struct thread_data *);
590extern void update_io_ticks(void);
591extern void disk_util_timer_arm(void);
592extern void setup_log(struct io_log **);
593extern void finish_log(struct thread_data *, struct io_log *, const char *);
594extern void __finish_log(struct io_log *, const char *);
595extern int setup_rate(struct thread_data *);
596extern struct io_log *agg_io_log[2];
597extern int write_bw_log;
598extern void add_agg_sample(unsigned long, enum fio_ddir);
599
600/*
601 * Time functions
602 */
603extern unsigned long utime_since(struct timeval *, struct timeval *);
604extern unsigned long utime_since_now(struct timeval *);
605extern unsigned long mtime_since(struct timeval *, struct timeval *);
606extern unsigned long mtime_since_now(struct timeval *);
607extern unsigned long time_since_now(struct timeval *);
608extern unsigned long mtime_since_genesis(void);
609extern void __usec_sleep(unsigned int);
610extern void usec_sleep(struct thread_data *, unsigned long);
611extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
612extern void fill_start_time(struct timeval *);
613extern void fio_gettime(struct timeval *, void *);
614extern void set_genesis_time(void);
615
616/*
617 * Init functions
618 */
619extern int __must_check parse_options(int, char **);
620extern int __must_check init_random_state(struct thread_data *);
621
622/*
623 * File setup/shutdown
624 */
625extern void close_files(struct thread_data *);
626extern int __must_check setup_files(struct thread_data *);
627extern int __must_check open_files(struct thread_data *);
628extern int open_file(struct thread_data *, struct fio_file *, int, int);
629extern void close_file(struct thread_data *, struct fio_file *);
630extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
631
632/*
633 * ETA/status stuff
634 */
635extern void print_thread_status(void);
636extern void print_status_init(int);
637
638/*
639 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
640 * will never back again. It may cycle between running/verififying/fsyncing.
641 * Once the thread reaches TD_EXITED, it is just waiting for the core to
642 * reap it.
643 */
644enum {
645 TD_NOT_CREATED = 0,
646 TD_CREATED,
647 TD_INITIALIZED,
648 TD_RUNNING,
649 TD_VERIFYING,
650 TD_FSYNCING,
651 TD_EXITED,
652 TD_REAPED,
653};
654
655/*
656 * Verify helpers
657 */
658extern void populate_verify_io_u(struct thread_data *, struct io_u *);
659extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
660extern int __must_check verify_io_u(struct io_u *);
661
662/*
663 * Memory helpers
664 */
665extern int __must_check fio_pin_memory(void);
666extern void fio_unpin_memory(void);
667extern int __must_check allocate_io_mem(struct thread_data *);
668extern void free_io_mem(struct thread_data *);
669
670/*
671 * io unit handling
672 */
673#define queue_full(td) list_empty(&(td)->io_u_freelist)
674extern struct io_u *__get_io_u(struct thread_data *);
675extern struct io_u *get_io_u(struct thread_data *);
676extern void put_io_u(struct thread_data *, struct io_u *);
677extern void requeue_io_u(struct thread_data *, struct io_u **);
678extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
679extern long __must_check io_u_queued_complete(struct thread_data *, int);
680extern void io_u_queued(struct thread_data *, struct io_u *);
681extern void io_u_log_error(struct thread_data *, struct io_u *);
682extern void io_u_init_timeout(void);
683extern void io_u_set_timeout(struct thread_data *);
684
685/*
686 * io engine entry points
687 */
688extern int __must_check td_io_init(struct thread_data *);
689extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
690extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
691extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
692extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
693extern int __must_check td_io_commit(struct thread_data *);
694
695/*
696 * This is a pretty crappy semaphore implementation, but with the use that fio
697 * has (just signalling start/go conditions), it doesn't have to be better.
698 * Naturally this would not work for any type of contended semaphore or
699 * for real locking.
700 */
701static inline void fio_sem_init(volatile int *sem, int val)
702{
703 *sem = val;
704}
705
706static inline void fio_sem_down(volatile int *sem)
707{
708 while (*sem == 0)
709 usleep(10000);
710
711 (*sem)--;
712}
713
714static inline void fio_sem_up(volatile int *sem)
715{
716 (*sem)++;
717}
718
719/*
720 * If logging output to a file, stderr should go to both stderr and f_err
721 */
722#define log_err(args...) do { \
723 fprintf(f_err, ##args); \
724 if (f_err != stderr) \
725 fprintf(stderr, ##args); \
726 } while (0)
727
728FILE *get_f_out(void);
729FILE *get_f_err(void);
730
731struct ioengine_ops {
732 struct list_head list;
733 char name[16];
734 int version;
735 int flags;
736 int (*setup)(struct thread_data *);
737 int (*init)(struct thread_data *);
738 int (*prep)(struct thread_data *, struct io_u *);
739 int (*queue)(struct thread_data *, struct io_u *);
740 int (*commit)(struct thread_data *);
741 int (*getevents)(struct thread_data *, int, int, struct timespec *);
742 struct io_u *(*event)(struct thread_data *, int);
743 int (*cancel)(struct thread_data *, struct io_u *);
744 void (*cleanup)(struct thread_data *);
745 void *data;
746 void *dlhandle;
747 unsigned long priv;
748};
749
750#define FIO_IOOPS_VERSION 5
751
752extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
753extern void register_ioengine(struct ioengine_ops *);
754extern void unregister_ioengine(struct ioengine_ops *);
755extern void close_ioengine(struct thread_data *);
756
757/*
758 * Mark unused variables passed to ops functions as unused, to silence gcc
759 */
760#define fio_unused __attribute((__unused__))
761#define fio_init __attribute__((constructor))
762#define fio_exit __attribute__((destructor))
763
764#define for_each_td(td, i) \
765 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
766#define for_each_file(td, f, i) \
767 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->nr_files; (i)++, (f)++)
768
769#define fio_assert(td, cond) do { \
770 if (!(cond)) { \
771 int *__foo = NULL; \
772 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
773 (td)->runstate = TD_EXITED; \
774 (td)->error = EFAULT; \
775 *__foo = 0; \
776 } \
777} while (0)
778
779#endif
fio - Flexible IO Tester