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