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