Fix bug with rwmix and multiple loops over the same set
[fio.git] / fio.h
CommitLineData
ebac4655
JA
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>
3c39a379
JA
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
6d6f031f 12#include <unistd.h>
34cfcdaf 13#include <string.h>
214e1eca 14#include <getopt.h>
cd14cc10 15#include <inttypes.h>
7101d9c2 16#include <assert.h>
ebac4655 17
317b95d0 18#include "compiler/compiler.h"
01743ee1 19#include "flist.h"
e2887563 20#include "fifo.h"
4b87898e 21#include "rbtree.h"
317b95d0
JA
22#include "arch/arch.h"
23#include "os/os.h"
07739b57 24#include "mutex.h"
a3d741fa
JA
25#include "log.h"
26#include "debug.h"
d6aed795
JA
27#include "file.h"
28#include "io_ddir.h"
dcefb588 29#include "ioengine.h"
5995a6a4 30#include "iolog.h"
ebac4655 31
609342ff
DL
32#ifdef FIO_HAVE_GUASI
33#include <guasi.h>
34#endif
35
417f0068
JA
36#ifdef FIO_HAVE_SOLARISAIO
37#include <sys/asynch.h>
38#endif
39
ebac4655 40struct group_run_stats {
9104f874
JA
41 unsigned long long max_run[2], min_run[2];
42 unsigned long long max_bw[2], min_bw[2];
e9b2a3fa 43 unsigned long long io_kb[2];
9104f874 44 unsigned long long agg[2];
ebac4655
JA
45};
46
e9c047a0
JA
47/*
48 * What type of allocation to use for io buffers
49 */
50enum fio_memtype {
51 MEM_MALLOC = 0, /* ordinary malloc */
52 MEM_SHM, /* use shared memory segments */
74b025b0 53 MEM_SHMHUGE, /* use shared memory segments with huge pages */
e9c047a0 54 MEM_MMAP, /* use anonynomous mmap */
d0bdaf49 55 MEM_MMAPHUGE, /* memory mapped huge file */
e9c047a0
JA
56};
57
b2560f3c
JA
58/*
59 * How many depth levels to log
60 */
61#define FIO_IO_U_MAP_NR 8
a9a18a3f
JA
62#define FIO_IO_U_LAT_U_NR 10
63#define FIO_IO_U_LAT_M_NR 12
b2560f3c 64
079ad09b 65struct thread_stat {
756867bd
JA
66 char *name;
67 char *verror;
68 int error;
69 int groupid;
70 pid_t pid;
71 char *description;
6586ee89 72 int members;
756867bd 73
079ad09b
JA
74 struct io_log *slat_log;
75 struct io_log *clat_log;
76 struct io_log *bw_log;
77
78 /*
79 * bandwidth and latency stats
80 */
81 struct io_stat clat_stat[2]; /* completion latency */
82 struct io_stat slat_stat[2]; /* submission latency */
83 struct io_stat bw_stat[2]; /* bandwidth stats */
84
85 unsigned long long stat_io_bytes[2];
86 struct timeval stat_sample_time[2];
87
88 /*
89 * fio system usage accounting
90 */
91 struct rusage ru_start;
92 struct rusage ru_end;
93 unsigned long usr_time;
94 unsigned long sys_time;
95 unsigned long ctx;
81887d5d 96 unsigned long minf, majf;
079ad09b 97
b2560f3c
JA
98 /*
99 * IO depth and latency stats
100 */
101 unsigned int io_u_map[FIO_IO_U_MAP_NR];
838bc709
JA
102 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
103 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
04a0feae
JA
104 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
105 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
b3605062 106 unsigned long total_io_u[2];
30061b97 107 unsigned long short_io_u[2];
838bc709
JA
108 unsigned long total_submit;
109 unsigned long total_complete;
756867bd
JA
110
111 unsigned long long io_bytes[2];
112 unsigned long runtime[2];
113 unsigned long total_run_time;
b2560f3c 114};
71619dc2 115
564ca972
JA
116struct bssplit {
117 unsigned int bs;
118 unsigned char perc;
119};
120
2dc1bbeb 121struct thread_options {
b1ec1da6 122 int pad;
61697c37 123 char *description;
b4692828 124 char *name;
ef899b63 125 char *directory;
13f8e2d2 126 char *filename;
2dc1bbeb 127 char *opendir;
09629a90 128 char *ioengine;
413dd459 129 enum td_ddir td_ddir;
211097b2 130 unsigned int ddir_nr;
2dc1bbeb
JA
131 unsigned int iodepth;
132 unsigned int iodepth_low;
133 unsigned int iodepth_batch;
4950421a 134 unsigned int iodepth_batch_complete;
2dc1bbeb
JA
135
136 unsigned long long size;
aa31f1f1 137 unsigned int fill_device;
2dc1bbeb
JA
138 unsigned long long file_size_low;
139 unsigned long long file_size_high;
140 unsigned long long start_offset;
141
142 unsigned int bs[2];
2b7a01d0 143 unsigned int ba[2];
2dc1bbeb
JA
144 unsigned int min_bs[2];
145 unsigned int max_bs[2];
720e84ad
JA
146 struct bssplit *bssplit[2];
147 unsigned int bssplit_nr[2];
2dc1bbeb
JA
148
149 unsigned int nr_files;
150 unsigned int open_files;
4d4e80f2 151 enum file_lock_mode file_lock_mode;
29c1349f 152 unsigned int lockfile_batch;
e9c047a0 153
9158d2f7
JA
154 unsigned int odirect;
155 unsigned int invalidate_cache;
156 unsigned int create_serialize;
157 unsigned int create_fsync;
814452bd 158 unsigned int create_on_open;
9158d2f7 159 unsigned int end_fsync;
afad68f7 160 unsigned int pre_read;
9158d2f7
JA
161 unsigned int sync_io;
162 unsigned int verify;
e84c73a8 163 unsigned int do_verify;
160b966d 164 unsigned int verifysort;
a59e170d
JA
165 unsigned int verify_interval;
166 unsigned int verify_offset;
90059d65
JA
167 unsigned int verify_pattern;
168 unsigned int verify_pattern_bytes;
a12a3b4d 169 unsigned int verify_fatal;
9158d2f7
JA
170 unsigned int use_thread;
171 unsigned int unlink;
172 unsigned int do_disk_util;
173 unsigned int override_sync;
174 unsigned int rand_repeatable;
175 unsigned int write_lat_log;
176 unsigned int write_bw_log;
bb8895e0 177 unsigned int norandommap;
2b386d25 178 unsigned int softrandommap;
690adba3 179 unsigned int bs_unaligned;
ebb1415f 180 unsigned int fsync_on_close;
e9c047a0 181
56bb17f2 182 unsigned int hugepage_size;
a00735e6 183 unsigned int rw_min_bs;
ebac4655 184 unsigned int thinktime;
48097d5c 185 unsigned int thinktime_spin;
9c1f7434 186 unsigned int thinktime_blocks;
ebac4655
JA
187 unsigned int fsync_blocks;
188 unsigned int start_delay;
0db26797 189 unsigned long long timeout;
721938ae 190 unsigned long long ramp_time;
ebac4655 191 unsigned int overwrite;
ebac4655 192 unsigned int bw_avg_time;
ebac4655 193 unsigned int loops;
20dc95c4
JA
194 unsigned long long zone_size;
195 unsigned long long zone_skip;
e9c047a0 196 enum fio_memtype mem_type;
2dc1bbeb 197
ebac4655 198 unsigned int stonewall;
b3d62a75 199 unsigned int new_group;
ebac4655 200 unsigned int numjobs;
ebac4655 201 os_cpu_mask_t cpumask;
375b2695 202 unsigned int cpumask_set;
aea47d44 203 unsigned int iolog;
a6ccc7be 204 unsigned int rwmixcycle;
e47f799f 205 unsigned int rwmix[2];
b6f4d880 206 unsigned int nice;
0aabe160 207 unsigned int file_service_type;
b2560f3c 208 unsigned int group_reporting;
d2f3ac35 209 unsigned int fadvise_hint;
e9459e5a 210 unsigned int zero_buffers;
5973cafb 211 unsigned int refill_buffers;
cf4464ca 212 unsigned int time_based;
9520ebb9
JA
213 unsigned int disable_clat;
214 unsigned int disable_slat;
215 unsigned int disable_bw;
993bf48b 216 unsigned int gtod_reduce;
be4ecfdf
JA
217 unsigned int gtod_cpu;
218 unsigned int gtod_offload;
aea47d44 219
076efc7c
JA
220 char *read_iolog_file;
221 char *write_iolog_file;
e3cedca7
JA
222 char *bw_log_file;
223 char *lat_log_file;
2dc1bbeb
JA
224
225 /*
226 * Pre-run and post-run shell
227 */
228 char *exec_prerun;
229 char *exec_postrun;
230
231 unsigned int rate;
232 unsigned int ratemin;
233 unsigned int ratecycle;
234 unsigned int rate_iops;
235 unsigned int rate_iops_min;
236
237 char *ioscheduler;
238
239 /*
240 * CPU "io" cycle burner
241 */
242 unsigned int cpuload;
243 unsigned int cpucycle;
244};
245
e4e33258
JA
246#define FIO_VERROR_SIZE 128
247
2dc1bbeb
JA
248/*
249 * This describes a single thread/process executing a fio job.
250 */
251struct thread_data {
252 struct thread_options o;
e4e33258 253 char verror[FIO_VERROR_SIZE];
2dc1bbeb
JA
254 pthread_t thread;
255 int thread_number;
256 int groupid;
257 struct thread_stat ts;
126d65c6 258 struct fio_file **files;
dd87b2c9 259 unsigned int files_size;
2dc1bbeb
JA
260 unsigned int files_index;
261 unsigned int nr_open_files;
1020a139 262 unsigned int nr_done_files;
2dc1bbeb
JA
263 unsigned int nr_normal_files;
264 union {
265 unsigned int next_file;
266 os_random_state_t next_file_state;
267 };
268 int error;
20e354ef 269 int done;
2dc1bbeb
JA
270 pid_t pid;
271 char *orig_buffer;
272 size_t orig_buffer_size;
273 volatile int terminate;
274 volatile int runstate;
275 unsigned int ioprio;
ac684785 276 unsigned int ioprio_set;
2dc1bbeb
JA
277 unsigned int last_was_sync;
278
279 char *mmapfile;
280 int mmapfd;
281
843a7413
JA
282 void *iolog_buf;
283 FILE *iolog_f;
ebac4655 284
da86774e 285 char *sysfs_root;
da86774e 286
5bfc35d7
JA
287 unsigned long rand_seeds[6];
288
6dfd46b9
JA
289 os_random_state_t bsrange_state;
290 os_random_state_t verify_state;
ebac4655
JA
291
292 int shm_id;
293
e9c047a0
JA
294 /*
295 * IO engine hooks, contains everything needed to submit an io_u
296 * to any of the available IO engines.
297 */
2866c82d 298 struct ioengine_ops *io_ops;
ebac4655 299
e9c047a0
JA
300 /*
301 * Current IO depth and list of free and busy io_u's.
302 */
ebac4655 303 unsigned int cur_depth;
d8005759 304 unsigned int io_u_queued;
01743ee1
JA
305 struct flist_head io_u_freelist;
306 struct flist_head io_u_busylist;
307 struct flist_head io_u_requeues;
ebac4655 308
e9c047a0
JA
309 /*
310 * Rate state
311 */
ebac4655
JA
312 unsigned long rate_usec_cycle;
313 long rate_pending_usleep;
314 unsigned long rate_bytes;
4e991c23 315 unsigned long rate_blocks;
ebac4655
JA
316 struct timeval lastrate;
317
ebac4655
JA
318 unsigned long long total_io_size;
319
755200a3 320 unsigned long io_issues[2];
9104f874
JA
321 unsigned long long io_blocks[2];
322 unsigned long long io_bytes[2];
48f5abd3 323 unsigned long long io_skip_bytes;
9104f874 324 unsigned long long this_io_bytes[2];
079ad09b 325 unsigned long long zone_bytes;
cdd18ad8 326 struct fio_mutex *mutex;
ebac4655 327
e9c047a0
JA
328 /*
329 * State for random io, a bitmap of blocks done vs not done
330 */
6dfd46b9 331 os_random_state_t random_state;
ebac4655 332
ebac4655
JA
333 struct timeval start; /* start of this loop */
334 struct timeval epoch; /* time job was started */
a61eddec 335 struct timeval last_issue;
993bf48b
JA
336 struct timeval tv_cache;
337 unsigned int tv_cache_nr;
338 unsigned int tv_cache_mask;
721938ae 339 unsigned int ramp_time_over;
ebac4655 340
e9c047a0
JA
341 /*
342 * read/write mixed workload state
343 */
6dfd46b9 344 os_random_state_t rwmix_state;
e4928662 345 unsigned long rwmix_issues;
e9c047a0 346 enum fio_ddir rwmix_ddir;
211097b2 347 unsigned int ddir_nr;
a6ccc7be 348
e9c047a0 349 /*
8de8f047
JA
350 * IO history logs for verification. We use a tree for sorting,
351 * if we are overwriting. Otherwise just use a fifo.
e9c047a0 352 */
4b87898e 353 struct rb_root io_hist_tree;
01743ee1 354 struct flist_head io_hist_list;
8de8f047
JA
355
356 /*
357 * For IO replaying
358 */
01743ee1 359 struct flist_head io_log_list;
433afcb4 360
1907dbc6
JA
361 /*
362 * for fileservice, how often to switch to a new file
363 */
364 unsigned int file_service_nr;
365 unsigned int file_service_left;
366 struct fio_file *file_service_file;
9c60ce64
JA
367
368 /*
369 * For generating file sizes
370 */
371 os_random_state_t file_size_state;
ebac4655
JA
372};
373
e592a06b
AC
374/*
375 * when should interactive ETA output be generated
376 */
377enum {
378 FIO_ETA_AUTO,
379 FIO_ETA_ALWAYS,
380 FIO_ETA_NEVER,
381};
382
e1161c32 383#define __td_verror(td, err, msg, func) \
ebac4655 384 do { \
19abcd3d
JA
385 if ((td)->error) \
386 break; \
ebac4655
JA
387 int e = (err); \
388 (td)->error = e; \
e1161c32 389 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
ebac4655
JA
390 } while (0)
391
b990b5c0 392
e1161c32
JA
393#define td_verror(td, err, func) \
394 __td_verror((td), (err), strerror((err)), (func))
395#define td_vmsg(td, err, msg, func) \
396 __td_verror((td), (err), (msg), (func))
b990b5c0 397
ebac4655
JA
398extern int exitall_on_terminate;
399extern int thread_number;
9cedf167 400extern int nr_process, nr_thread;
ebac4655
JA
401extern int shm_id;
402extern int groupid;
c6ae0a5b 403extern int terse_output;
53cdc686 404extern int temp_stall_ts;
1e97cce9 405extern unsigned long long mlock_size;
cfc99db7 406extern unsigned long page_mask, page_size;
4241ea8f 407extern int read_only;
e592a06b 408extern int eta_print;
d3eeeabc 409extern unsigned long done_secs;
01f06b63 410extern char *job_section;
be4ecfdf
JA
411extern int fio_gtod_offload;
412extern int fio_gtod_cpu;
ebac4655
JA
413
414extern struct thread_data *threads;
415
2dc1bbeb
JA
416#define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
417#define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
418#define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
419#define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
303032ae 420#define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
ebac4655 421
7101d9c2
JA
422static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
423{
424 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
425}
426
de605666 427#define BLOCKS_PER_MAP (8 * sizeof(int))
aec2de20
JA
428#define TO_MAP_BLOCK(f, b) (b)
429#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
430#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
ebac4655
JA
431
432#define MAX_JOBS (1024)
433
87dc1ab1
JA
434static inline int should_fsync(struct thread_data *td)
435{
436 if (td->last_was_sync)
437 return 0;
2dc1bbeb 438 if (td->o.odirect)
87dc1ab1 439 return 0;
2dc1bbeb 440 if (td_write(td) || td_rw(td) || td->o.override_sync)
87dc1ab1
JA
441 return 1;
442
443 return 0;
444}
445
6796209a
JA
446/*
447 * Time functions
448 */
10927316
SL
449extern unsigned long long utime_since(struct timeval *, struct timeval *);
450extern unsigned long long utime_since_now(struct timeval *);
6796209a
JA
451extern unsigned long mtime_since(struct timeval *, struct timeval *);
452extern unsigned long mtime_since_now(struct timeval *);
453extern unsigned long time_since_now(struct timeval *);
263e529f 454extern unsigned long mtime_since_genesis(void);
6dd6f2cd 455extern void usec_spin(unsigned int);
6796209a 456extern void usec_sleep(struct thread_data *, unsigned long);
413dd459 457extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
6043c579 458extern void fill_start_time(struct timeval *);
02bcaa8c 459extern void fio_gettime(struct timeval *, void *);
be4ecfdf
JA
460extern void fio_gtod_init(void);
461extern void fio_gtod_update(void);
a2f77c9f 462extern void set_genesis_time(void);
721938ae 463extern int ramp_time_over(struct thread_data *);
b29ee5b3 464extern int in_ramp_time(struct thread_data *);
6796209a 465
8914a9d8 466/*
214e1eca 467 * Init/option functions
8914a9d8 468 */
72cb971b 469extern int __must_check parse_options(int, char **);
3b8b7135 470extern int fio_options_parse(struct thread_data *, char **, int);
214e1eca
JA
471extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
472extern void fio_fill_default_options(struct thread_data *);
473extern int fio_show_option_help(const char *);
474extern void fio_options_dup_and_init(struct option *);
d23bb327
JA
475extern void options_mem_dupe(struct thread_data *);
476extern void options_mem_free(struct thread_data *);
5bfc35d7 477extern void td_fill_rand_seeds(struct thread_data *);
214e1eca
JA
478#define FIO_GETOPT_JOB 0x89988998
479#define FIO_NR_OPTIONS 128
8914a9d8 480
263e529f
JA
481/*
482 * ETA/status stuff
483 */
484extern void print_thread_status(void);
485extern void print_status_init(int);
486
487/*
488 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
489 * will never back again. It may cycle between running/verififying/fsyncing.
490 * Once the thread reaches TD_EXITED, it is just waiting for the core to
491 * reap it.
492 */
493enum {
494 TD_NOT_CREATED = 0,
495 TD_CREATED,
496 TD_INITIALIZED,
b29ee5b3 497 TD_RAMP,
263e529f 498 TD_RUNNING,
b0f65863 499 TD_PRE_READING,
263e529f
JA
500 TD_VERIFYING,
501 TD_FSYNCING,
502 TD_EXITED,
503 TD_REAPED,
504};
505
b29ee5b3
JA
506extern void td_set_runstate(struct thread_data *, int);
507
2f9ade3c
JA
508/*
509 * Memory helpers
510 */
b2fdda43 511extern int __must_check fio_pin_memory(void);
2f9ade3c 512extern void fio_unpin_memory(void);
b2fdda43 513extern int __must_check allocate_io_mem(struct thread_data *);
2f9ade3c
JA
514extern void free_io_mem(struct thread_data *);
515
b29ee5b3
JA
516/*
517 * Reset stats after ramp time completes
518 */
519extern void reset_all_stats(struct thread_data *);
520
fb7b71a3
JA
521/*
522 * blktrace support
523 */
5e62c22a 524#ifdef FIO_HAVE_BLKTRACE
fb7b71a3
JA
525extern int is_blktrace(const char *);
526extern int load_blktrace(struct thread_data *, const char *);
5e62c22a 527#endif
fb7b71a3 528
2866c82d
JA
529/*
530 * Mark unused variables passed to ops functions as unused, to silence gcc
531 */
532#define fio_unused __attribute((__unused__))
5f350952
JA
533#define fio_init __attribute__((constructor))
534#define fio_exit __attribute__((destructor))
2866c82d 535
34572e28
JA
536#define for_each_td(td, i) \
537 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686 538#define for_each_file(td, f, i) \
691c8fb0
JA
539 if ((td)->files_index) \
540 for ((i) = 0, (f) = (td)->files[0]; \
541 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
542 (i)++)
53cdc686 543
0032bf9f
JA
544#define fio_assert(td, cond) do { \
545 if (!(cond)) { \
340fd243 546 int *__foo = NULL; \
0032bf9f
JA
547 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
548 (td)->runstate = TD_EXITED; \
437c9b71 549 (td)->error = EFAULT; \
340fd243 550 *__foo = 0; \
0032bf9f
JA
551 } \
552} while (0)
553
12d9d841
JA
554static inline int fio_fill_issue_time(struct thread_data *td)
555{
556 if (td->o.read_iolog_file ||
557 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
558 return 1;
559
560 return 0;
561}
562
ebac4655 563#endif