projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
man page typo
[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#include <inttypes.h>
15#include <assert.h>
16
17struct thread_data;
18
19#include "compiler/compiler.h"
20#include "flist.h"
21#include "fifo.h"
22#include "rbtree.h"
23#include "arch/arch.h"
24#include "os/os.h"
25#include "mutex.h"
26#include "log.h"
27#include "debug.h"
28#include "file.h"
29#include "io_ddir.h"
30#include "ioengine.h"
31#include "iolog.h"
32#include "helpers.h"
33#include "options.h"
34#include "profile.h"
35#include "time.h"
36#include "lib/getopt.h"
37#include "lib/rand.h"
38
39#ifdef FIO_HAVE_GUASI
40#include <guasi.h>
41#endif
42
43#ifdef FIO_HAVE_SOLARISAIO
44#include <sys/asynch.h>
45#endif
46
47struct group_run_stats {
48 unsigned long long max_run[2], min_run[2];
49 unsigned long long max_bw[2], min_bw[2];
50 unsigned long long io_kb[2];
51 unsigned long long agg[2];
52 unsigned int kb_base;
53};
54
55/*
56 * What type of allocation to use for io buffers
57 */
58enum fio_memtype {
59 MEM_MALLOC = 0, /* ordinary malloc */
60 MEM_SHM, /* use shared memory segments */
61 MEM_SHMHUGE, /* use shared memory segments with huge pages */
62 MEM_MMAP, /* use anonynomous mmap */
63 MEM_MMAPHUGE, /* memory mapped huge file */
64};
65
66/*
67 * offset generator types
68 */
69enum {
70 RW_SEQ_SEQ = 0,
71 RW_SEQ_IDENT,
72};
73
74/*
75 * How many depth levels to log
76 */
77#define FIO_IO_U_MAP_NR 7
78#define FIO_IO_U_LAT_U_NR 10
79#define FIO_IO_U_LAT_M_NR 12
80
81#define MAX_PATTERN_SIZE 512
82
83struct thread_stat {
84 char *name;
85 char *verror;
86 int error;
87 int groupid;
88 pid_t pid;
89 char *description;
90 int members;
91
92 struct io_log *slat_log;
93 struct io_log *clat_log;
94 struct io_log *lat_log;
95 struct io_log *bw_log;
96
97 /*
98 * bandwidth and latency stats
99 */
100 struct io_stat clat_stat[2]; /* completion latency */
101 struct io_stat slat_stat[2]; /* submission latency */
102 struct io_stat lat_stat[2]; /* total latency */
103 struct io_stat bw_stat[2]; /* bandwidth stats */
104
105 unsigned long long stat_io_bytes[2];
106 struct timeval stat_sample_time[2];
107
108 /*
109 * fio system usage accounting
110 */
111 struct rusage ru_start;
112 struct rusage ru_end;
113 unsigned long usr_time;
114 unsigned long sys_time;
115 unsigned long ctx;
116 unsigned long minf, majf;
117
118 /*
119 * IO depth and latency stats
120 */
121 unsigned int io_u_map[FIO_IO_U_MAP_NR];
122 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
123 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
124 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
125 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
126 unsigned long total_io_u[3];
127 unsigned long short_io_u[3];
128 unsigned long total_submit;
129 unsigned long total_complete;
130
131 unsigned long long io_bytes[2];
132 unsigned long long runtime[2];
133 unsigned long total_run_time;
134
135 /*
136 * IO Error related stats
137 */
138 unsigned continue_on_error;
139 unsigned long total_err_count;
140 int first_error;
141
142 unsigned int kb_base;
143};
144
145struct bssplit {
146 unsigned int bs;
147 unsigned char perc;
148};
149
150struct thread_options {
151 int pad;
152 char *description;
153 char *name;
154 char *directory;
155 char *filename;
156 char *opendir;
157 char *ioengine;
158 enum td_ddir td_ddir;
159 unsigned int rw_seq;
160 unsigned int kb_base;
161 unsigned int ddir_seq_nr;
162 unsigned int iodepth;
163 unsigned int iodepth_low;
164 unsigned int iodepth_batch;
165 unsigned int iodepth_batch_complete;
166
167 unsigned long long size;
168 unsigned int size_percent;
169 unsigned int fill_device;
170 unsigned long long file_size_low;
171 unsigned long long file_size_high;
172 unsigned long long start_offset;
173
174 unsigned int bs[2];
175 unsigned int ba[2];
176 unsigned int min_bs[2];
177 unsigned int max_bs[2];
178 struct bssplit *bssplit[2];
179 unsigned int bssplit_nr[2];
180
181 unsigned int nr_files;
182 unsigned int open_files;
183 enum file_lock_mode file_lock_mode;
184 unsigned int lockfile_batch;
185
186 unsigned int odirect;
187 unsigned int invalidate_cache;
188 unsigned int create_serialize;
189 unsigned int create_fsync;
190 unsigned int create_on_open;
191 unsigned int end_fsync;
192 unsigned int pre_read;
193 unsigned int sync_io;
194 unsigned int verify;
195 unsigned int do_verify;
196 unsigned int verifysort;
197 unsigned int verify_interval;
198 unsigned int verify_offset;
199 char verify_pattern[MAX_PATTERN_SIZE];
200 unsigned int verify_pattern_bytes;
201 unsigned int verify_fatal;
202 unsigned int verify_dump;
203 unsigned int verify_async;
204 unsigned long long verify_backlog;
205 unsigned int verify_batch;
206 unsigned int use_thread;
207 unsigned int unlink;
208 unsigned int do_disk_util;
209 unsigned int override_sync;
210 unsigned int rand_repeatable;
211 unsigned int use_os_rand;
212 unsigned int write_lat_log;
213 unsigned int write_bw_log;
214 unsigned int norandommap;
215 unsigned int softrandommap;
216 unsigned int bs_unaligned;
217 unsigned int fsync_on_close;
218
219 unsigned int hugepage_size;
220 unsigned int rw_min_bs;
221 unsigned int thinktime;
222 unsigned int thinktime_spin;
223 unsigned int thinktime_blocks;
224 unsigned int fsync_blocks;
225 unsigned int fdatasync_blocks;
226 unsigned int barrier_blocks;
227 unsigned long long start_delay;
228 unsigned long long timeout;
229 unsigned long long ramp_time;
230 unsigned int overwrite;
231 unsigned int bw_avg_time;
232 unsigned int loops;
233 unsigned long long zone_size;
234 unsigned long long zone_skip;
235 enum fio_memtype mem_type;
236 unsigned int mem_align;
237
238 unsigned int stonewall;
239 unsigned int new_group;
240 unsigned int numjobs;
241 os_cpu_mask_t cpumask;
242 unsigned int cpumask_set;
243 os_cpu_mask_t verify_cpumask;
244 unsigned int verify_cpumask_set;
245 unsigned int iolog;
246 unsigned int rwmixcycle;
247 unsigned int rwmix[2];
248 unsigned int nice;
249 unsigned int file_service_type;
250 unsigned int group_reporting;
251 unsigned int fadvise_hint;
252 enum fio_fallocate_mode fallocate_mode;
253 unsigned int zero_buffers;
254 unsigned int refill_buffers;
255 unsigned int time_based;
256 unsigned int disable_lat;
257 unsigned int disable_clat;
258 unsigned int disable_slat;
259 unsigned int disable_bw;
260 unsigned int gtod_reduce;
261 unsigned int gtod_cpu;
262 unsigned int gtod_offload;
263 enum fio_cs clocksource;
264 unsigned int no_stall;
265 unsigned int trim_percentage;
266 unsigned int trim_batch;
267 unsigned int trim_zero;
268 unsigned long long trim_backlog;
269
270 char *read_iolog_file;
271 char *write_iolog_file;
272 char *bw_log_file;
273 char *lat_log_file;
274 char *replay_redirect;
275
276 /*
277 * Pre-run and post-run shell
278 */
279 char *exec_prerun;
280 char *exec_postrun;
281
282 unsigned int rate[2];
283 unsigned int ratemin[2];
284 unsigned int ratecycle;
285 unsigned int rate_iops[2];
286 unsigned int rate_iops_min[2];
287
288 char *ioscheduler;
289
290 /*
291 * CPU "io" cycle burner
292 */
293 unsigned int cpuload;
294 unsigned int cpucycle;
295
296 /*
297 * I/O Error handling
298 */
299 unsigned int continue_on_error;
300
301 /*
302 * Benchmark profile type
303 */
304 char *profile;
305
306 /*
307 * blkio cgroup support
308 */
309 char *cgroup;
310 unsigned int cgroup_weight;
311 unsigned int cgroup_nodelete;
312
313 unsigned int uid;
314 unsigned int gid;
315
316 unsigned int sync_file_range;
317};
318
319#define FIO_VERROR_SIZE 128
320
321/*
322 * This describes a single thread/process executing a fio job.
323 */
324struct thread_data {
325 struct thread_options o;
326 char verror[FIO_VERROR_SIZE];
327 pthread_t thread;
328 int thread_number;
329 int groupid;
330 struct thread_stat ts;
331 struct fio_file **files;
332 unsigned int files_size;
333 unsigned int files_index;
334 unsigned int nr_open_files;
335 unsigned int nr_done_files;
336 unsigned int nr_normal_files;
337 union {
338 unsigned int next_file;
339 os_random_state_t next_file_state;
340 struct frand_state __next_file_state;
341 };
342 int error;
343 int done;
344 pid_t pid;
345 char *orig_buffer;
346 size_t orig_buffer_size;
347 volatile int terminate;
348 volatile int runstate;
349 unsigned int ioprio;
350 unsigned int ioprio_set;
351 unsigned int last_was_sync;
352 enum fio_ddir last_ddir;
353
354 char *mmapfile;
355 int mmapfd;
356
357 void *iolog_buf;
358 FILE *iolog_f;
359
360 char *sysfs_root;
361
362 unsigned long rand_seeds[7];
363
364 union {
365 os_random_state_t bsrange_state;
366 struct frand_state __bsrange_state;
367 };
368 union {
369 os_random_state_t verify_state;
370 struct frand_state __verify_state;
371 };
372 union {
373 os_random_state_t trim_state;
374 struct frand_state __trim_state;
375 };
376
377 unsigned int verify_batch;
378 unsigned int trim_batch;
379
380 int shm_id;
381
382 /*
383 * IO engine hooks, contains everything needed to submit an io_u
384 * to any of the available IO engines.
385 */
386 struct ioengine_ops *io_ops;
387
388 /*
389 * Current IO depth and list of free and busy io_u's.
390 */
391 unsigned int cur_depth;
392 unsigned int io_u_queued;
393 struct flist_head io_u_freelist;
394 struct flist_head io_u_busylist;
395 struct flist_head io_u_requeues;
396 pthread_mutex_t io_u_lock;
397 pthread_cond_t free_cond;
398
399 /*
400 * async verify offload
401 */
402 struct flist_head verify_list;
403 pthread_t *verify_threads;
404 unsigned int nr_verify_threads;
405 pthread_cond_t verify_cond;
406 int verify_thread_exit;
407
408 /*
409 * Rate state
410 */
411 unsigned long rate_nsec_cycle[2];
412 long rate_pending_usleep[2];
413 unsigned long rate_bytes[2];
414 unsigned long rate_blocks[2];
415 struct timeval lastrate[2];
416
417 unsigned long long total_io_size;
418 unsigned long long fill_device_size;
419
420 unsigned long io_issues[2];
421 unsigned long long io_blocks[2];
422 unsigned long long io_bytes[2];
423 unsigned long long io_skip_bytes;
424 unsigned long long this_io_bytes[2];
425 unsigned long long zone_bytes;
426 struct fio_mutex *mutex;
427
428 /*
429 * State for random io, a bitmap of blocks done vs not done
430 */
431 union {
432 os_random_state_t random_state;
433 struct frand_state __random_state;
434 };
435
436 struct timeval start; /* start of this loop */
437 struct timeval epoch; /* time job was started */
438 struct timeval last_issue;
439 struct timeval tv_cache;
440 unsigned int tv_cache_nr;
441 unsigned int tv_cache_mask;
442 unsigned int ramp_time_over;
443
444 /*
445 * read/write mixed workload state
446 */
447 union {
448 os_random_state_t rwmix_state;
449 struct frand_state __rwmix_state;
450 };
451 unsigned long rwmix_issues;
452 enum fio_ddir rwmix_ddir;
453 unsigned int ddir_seq_nr;
454
455 /*
456 * IO history logs for verification. We use a tree for sorting,
457 * if we are overwriting. Otherwise just use a fifo.
458 */
459 struct rb_root io_hist_tree;
460 struct flist_head io_hist_list;
461 unsigned long io_hist_len;
462
463 /*
464 * For IO replaying
465 */
466 struct flist_head io_log_list;
467
468 /*
469 * For tracking/handling discards
470 */
471 struct flist_head trim_list;
472 unsigned long trim_entries;
473
474 /*
475 * for fileservice, how often to switch to a new file
476 */
477 unsigned int file_service_nr;
478 unsigned int file_service_left;
479 struct fio_file *file_service_file;
480
481 unsigned int sync_file_range_nr;
482
483 /*
484 * For generating file sizes
485 */
486 union {
487 os_random_state_t file_size_state;
488 struct frand_state __file_size_state;
489 };
490
491 /*
492 * Error counts
493 */
494 unsigned int total_err_count;
495 int first_error;
496
497 /*
498 * Can be overloaded by profiles
499 */
500 struct prof_io_ops prof_io_ops;
501 void *prof_data;
502};
503
504/*
505 * when should interactive ETA output be generated
506 */
507enum {
508 FIO_ETA_AUTO,
509 FIO_ETA_ALWAYS,
510 FIO_ETA_NEVER,
511};
512
513#define __td_verror(td, err, msg, func) \
514 do { \
515 if ((td)->error) \
516 break; \
517 int e = (err); \
518 (td)->error = e; \
519 if (!(td)->first_error) \
520 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
521 } while (0)
522
523
524#define td_clear_error(td) \
525 (td)->error = 0;
526#define td_verror(td, err, func) \
527 __td_verror((td), (err), strerror((err)), (func))
528#define td_vmsg(td, err, msg, func) \
529 __td_verror((td), (err), (msg), (func))
530
531extern int exitall_on_terminate;
532extern int thread_number;
533extern int nr_process, nr_thread;
534extern int shm_id;
535extern int groupid;
536extern int terse_output;
537extern int temp_stall_ts;
538extern unsigned long long mlock_size;
539extern unsigned long page_mask, page_size;
540extern int read_only;
541extern int eta_print;
542extern unsigned long done_secs;
543extern char *job_section;
544extern int fio_gtod_offload;
545extern int fio_gtod_cpu;
546extern enum fio_cs fio_clock_source;
547extern int warnings_fatal;
548
549extern struct thread_data *threads;
550
551static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
552{
553 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
554}
555
556#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
557#define TO_MAP_BLOCK(f, b) (b)
558#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
559#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
560
561#define REAL_MAX_JOBS 2048
562
563#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
564
565static inline void update_error_count(struct thread_data *td, int err)
566{
567 td->total_err_count++;
568 if (td->total_err_count == 1)
569 td->first_error = err;
570}
571
572static inline int should_fsync(struct thread_data *td)
573{
574 if (td->last_was_sync)
575 return 0;
576 if (td->o.odirect)
577 return 0;
578 if (td_write(td) || td_rw(td) || td->o.override_sync)
579 return 1;
580
581 return 0;
582}
583
584/*
585 * Init/option functions
586 */
587extern int __must_check parse_options(int, char **);
588extern int fio_options_parse(struct thread_data *, char **, int);
589extern void fio_keywords_init(void);
590extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
591extern void fio_fill_default_options(struct thread_data *);
592extern int fio_show_option_help(const char *);
593extern void fio_options_dup_and_init(struct option *);
594extern void options_mem_dupe(struct thread_data *);
595extern void options_mem_free(struct thread_data *);
596extern void td_fill_rand_seeds(struct thread_data *);
597extern void add_job_opts(const char **);
598extern char *num2str(unsigned long, int, int, int);
599
600#define FIO_GETOPT_JOB 0x89988998
601#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
602
603/*
604 * ETA/status stuff
605 */
606extern void print_thread_status(void);
607extern void print_status_init(int);
608
609/*
610 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
611 * will never back again. It may cycle between running/verififying/fsyncing.
612 * Once the thread reaches TD_EXITED, it is just waiting for the core to
613 * reap it.
614 */
615enum {
616 TD_NOT_CREATED = 0,
617 TD_CREATED,
618 TD_INITIALIZED,
619 TD_RAMP,
620 TD_RUNNING,
621 TD_PRE_READING,
622 TD_VERIFYING,
623 TD_FSYNCING,
624 TD_EXITED,
625 TD_REAPED,
626};
627
628extern void td_set_runstate(struct thread_data *, int);
629
630/*
631 * Memory helpers
632 */
633extern int __must_check fio_pin_memory(void);
634extern void fio_unpin_memory(void);
635extern int __must_check allocate_io_mem(struct thread_data *);
636extern void free_io_mem(struct thread_data *);
637
638/*
639 * Reset stats after ramp time completes
640 */
641extern void reset_all_stats(struct thread_data *);
642
643/*
644 * blktrace support
645 */
646#ifdef FIO_HAVE_BLKTRACE
647extern int is_blktrace(const char *);
648extern int load_blktrace(struct thread_data *, const char *);
649#endif
650
651/*
652 * Mark unused variables passed to ops functions as unused, to silence gcc
653 */
654#define fio_unused __attribute((__unused__))
655#define fio_init __attribute__((constructor))
656#define fio_exit __attribute__((destructor))
657
658#define for_each_td(td, i) \
659 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
660#define for_each_file(td, f, i) \
661 if ((td)->files_index) \
662 for ((i) = 0, (f) = (td)->files[0]; \
663 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
664 (i)++)
665
666#define fio_assert(td, cond) do { \
667 if (!(cond)) { \
668 int *__foo = NULL; \
669 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
670 td_set_runstate((td), TD_EXITED); \
671 (td)->error = EFAULT; \
672 *__foo = 0; \
673 } \
674} while (0)
675
676static inline int fio_fill_issue_time(struct thread_data *td)
677{
678 if (td->o.read_iolog_file ||
679 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
680 return 1;
681
682 return 0;
683}
684
685static inline int __should_check_rate(struct thread_data *td,
686 enum fio_ddir ddir)
687{
688 struct thread_options *o = &td->o;
689
690 /*
691 * If some rate setting was given, we need to check it
692 */
693 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
694 o->rate_iops_min[ddir])
695 return 1;
696
697 return 0;
698}
699
700static inline int should_check_rate(struct thread_data *td,
701 unsigned long *bytes_done)
702{
703 int ret = 0;
704
705 if (bytes_done[0])
706 ret |= __should_check_rate(td, 0);
707 if (bytes_done[1])
708 ret |= __should_check_rate(td, 1);
709
710 return ret;
711}
712
713static inline int is_power_of_2(unsigned int val)
714{
715 return (val != 0 && ((val & (val - 1)) == 0));
716}
717
718/*
719 * We currently only need to do locking if we have verifier threads
720 * accessing our internal structures too
721 */
722static inline void td_io_u_lock(struct thread_data *td)
723{
724 if (td->o.verify_async)
725 pthread_mutex_lock(&td->io_u_lock);
726}
727
728static inline void td_io_u_unlock(struct thread_data *td)
729{
730 if (td->o.verify_async)
731 pthread_mutex_unlock(&td->io_u_lock);
732}
733
734static inline void td_io_u_free_notify(struct thread_data *td)
735{
736 if (td->o.verify_async)
737 pthread_cond_signal(&td->free_cond);
738}
739
740#endif
fio - Flexible IO Tester