Get rid of fio_version.h
[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 "gettime.h"
37#include "lib/getopt.h"
38#include "lib/rand.h"
39#include "server.h"
40#include "stat.h"
41#include "flow.h"
42
43#ifdef FIO_HAVE_GUASI
44#include <guasi.h>
45#endif
46
47#ifdef FIO_HAVE_SOLARISAIO
48#include <sys/asynch.h>
49#endif
50
51/*
52 * What type of allocation to use for io buffers
53 */
54enum fio_memtype {
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
60};
61
62/*
63 * offset generator types
64 */
65enum {
66 RW_SEQ_SEQ = 0,
67 RW_SEQ_IDENT,
68};
69
70/*
71 * What type of errors to continue on when continue_on_error is used
72 */
73enum error_type {
74 ERROR_TYPE_NONE = 0,
75 ERROR_TYPE_READ = 1 << 0,
76 ERROR_TYPE_WRITE = 1 << 1,
77 ERROR_TYPE_VERIFY = 1 << 2,
78 ERROR_TYPE_ANY = 0xffff,
79};
80
81struct bssplit {
82 unsigned int bs;
83 unsigned char perc;
84};
85
86struct thread_options {
87 int pad;
88 char *description;
89 char *name;
90 char *directory;
91 char *filename;
92 char *opendir;
93 char *ioengine;
94 enum td_ddir td_ddir;
95 unsigned int rw_seq;
96 unsigned int kb_base;
97 unsigned int ddir_seq_nr;
98 long ddir_seq_add;
99 unsigned int iodepth;
100 unsigned int iodepth_low;
101 unsigned int iodepth_batch;
102 unsigned int iodepth_batch_complete;
103
104 unsigned long long size;
105 unsigned int size_percent;
106 unsigned int fill_device;
107 unsigned long long file_size_low;
108 unsigned long long file_size_high;
109 unsigned long long start_offset;
110
111 unsigned int bs[2];
112 unsigned int ba[2];
113 unsigned int min_bs[2];
114 unsigned int max_bs[2];
115 struct bssplit *bssplit[2];
116 unsigned int bssplit_nr[2];
117
118 unsigned int nr_files;
119 unsigned int open_files;
120 enum file_lock_mode file_lock_mode;
121 unsigned int lockfile_batch;
122
123 unsigned int odirect;
124 unsigned int invalidate_cache;
125 unsigned int create_serialize;
126 unsigned int create_fsync;
127 unsigned int create_on_open;
128 unsigned int end_fsync;
129 unsigned int pre_read;
130 unsigned int sync_io;
131 unsigned int verify;
132 unsigned int do_verify;
133 unsigned int verifysort;
134 unsigned int verify_interval;
135 unsigned int verify_offset;
136 char verify_pattern[MAX_PATTERN_SIZE];
137 unsigned int verify_pattern_bytes;
138 unsigned int verify_fatal;
139 unsigned int verify_dump;
140 unsigned int verify_async;
141 unsigned long long verify_backlog;
142 unsigned int verify_batch;
143 unsigned int use_thread;
144 unsigned int unlink;
145 unsigned int do_disk_util;
146 unsigned int override_sync;
147 unsigned int rand_repeatable;
148 unsigned int use_os_rand;
149 unsigned int write_lat_log;
150 unsigned int write_bw_log;
151 unsigned int write_iops_log;
152 unsigned int log_avg_msec;
153 unsigned int norandommap;
154 unsigned int softrandommap;
155 unsigned int bs_unaligned;
156 unsigned int fsync_on_close;
157
158 unsigned int hugepage_size;
159 unsigned int rw_min_bs;
160 unsigned int thinktime;
161 unsigned int thinktime_spin;
162 unsigned int thinktime_blocks;
163 unsigned int fsync_blocks;
164 unsigned int fdatasync_blocks;
165 unsigned int barrier_blocks;
166 unsigned long long start_delay;
167 unsigned long long timeout;
168 unsigned long long ramp_time;
169 unsigned int overwrite;
170 unsigned int bw_avg_time;
171 unsigned int iops_avg_time;
172 unsigned int loops;
173 unsigned long long zone_range;
174 unsigned long long zone_size;
175 unsigned long long zone_skip;
176 enum fio_memtype mem_type;
177 unsigned int mem_align;
178
179 unsigned int stonewall;
180 unsigned int new_group;
181 unsigned int numjobs;
182 os_cpu_mask_t cpumask;
183 unsigned int cpumask_set;
184 os_cpu_mask_t verify_cpumask;
185 unsigned int verify_cpumask_set;
186 unsigned int iolog;
187 unsigned int rwmixcycle;
188 unsigned int rwmix[2];
189 unsigned int nice;
190 unsigned int file_service_type;
191 unsigned int group_reporting;
192 unsigned int fadvise_hint;
193 enum fio_fallocate_mode fallocate_mode;
194 unsigned int zero_buffers;
195 unsigned int refill_buffers;
196 unsigned int scramble_buffers;
197 unsigned int compress_percentage;
198 unsigned int compress_chunk;
199 unsigned int time_based;
200 unsigned int disable_lat;
201 unsigned int disable_clat;
202 unsigned int disable_slat;
203 unsigned int disable_bw;
204 unsigned int gtod_reduce;
205 unsigned int gtod_cpu;
206 unsigned int gtod_offload;
207 enum fio_cs clocksource;
208 unsigned int no_stall;
209 unsigned int trim_percentage;
210 unsigned int trim_batch;
211 unsigned int trim_zero;
212 unsigned long long trim_backlog;
213 unsigned int clat_percentiles;
214 unsigned int overwrite_plist;
215 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
216
217 char *read_iolog_file;
218 char *write_iolog_file;
219 char *bw_log_file;
220 char *lat_log_file;
221 char *iops_log_file;
222 char *replay_redirect;
223
224 /*
225 * Pre-run and post-run shell
226 */
227 char *exec_prerun;
228 char *exec_postrun;
229
230 unsigned int rate[2];
231 unsigned int ratemin[2];
232 unsigned int ratecycle;
233 unsigned int rate_iops[2];
234 unsigned int rate_iops_min[2];
235
236 char *ioscheduler;
237
238 /*
239 * CPU "io" cycle burner
240 */
241 unsigned int cpuload;
242 unsigned int cpucycle;
243
244 /*
245 * I/O Error handling
246 */
247 enum error_type continue_on_error;
248
249 /*
250 * Benchmark profile type
251 */
252 char *profile;
253
254 /*
255 * blkio cgroup support
256 */
257 char *cgroup;
258 unsigned int cgroup_weight;
259 unsigned int cgroup_nodelete;
260
261 unsigned int uid;
262 unsigned int gid;
263
264 int flow_id;
265 int flow;
266 int flow_watermark;
267 unsigned int flow_sleep;
268
269 unsigned long long offset_increment;
270
271 unsigned int sync_file_range;
272};
273
274/*
275 * This describes a single thread/process executing a fio job.
276 */
277struct thread_data {
278 struct thread_options o;
279 void *eo;
280 char verror[FIO_VERROR_SIZE];
281 pthread_t thread;
282 int thread_number;
283 int groupid;
284 struct thread_stat ts;
285
286 struct io_log *slat_log;
287 struct io_log *clat_log;
288 struct io_log *lat_log;
289 struct io_log *bw_log;
290 struct io_log *iops_log;
291
292 uint64_t stat_io_bytes[2];
293 struct timeval bw_sample_time;
294
295 uint64_t stat_io_blocks[2];
296 struct timeval iops_sample_time;
297
298 struct rusage ru_start;
299 struct rusage ru_end;
300
301 struct fio_file **files;
302 unsigned int files_size;
303 unsigned int files_index;
304 unsigned int nr_open_files;
305 unsigned int nr_done_files;
306 unsigned int nr_normal_files;
307 union {
308 unsigned int next_file;
309 os_random_state_t next_file_state;
310 struct frand_state __next_file_state;
311 };
312 int error;
313 int sig;
314 int done;
315 pid_t pid;
316 char *orig_buffer;
317 size_t orig_buffer_size;
318 volatile int terminate;
319 volatile int runstate;
320 unsigned int ioprio;
321 unsigned int ioprio_set;
322 unsigned int last_was_sync;
323 enum fio_ddir last_ddir;
324
325 char *mmapfile;
326 int mmapfd;
327
328 void *iolog_buf;
329 FILE *iolog_f;
330
331 char *sysfs_root;
332
333 unsigned long rand_seeds[8];
334
335 union {
336 os_random_state_t bsrange_state;
337 struct frand_state __bsrange_state;
338 };
339 union {
340 os_random_state_t verify_state;
341 struct frand_state __verify_state;
342 };
343 union {
344 os_random_state_t trim_state;
345 struct frand_state __trim_state;
346 };
347
348 struct frand_state buf_state;
349
350 unsigned int verify_batch;
351 unsigned int trim_batch;
352
353 int shm_id;
354
355 /*
356 * IO engine hooks, contains everything needed to submit an io_u
357 * to any of the available IO engines.
358 */
359 struct ioengine_ops *io_ops;
360
361 /*
362 * Queue depth of io_u's that fio MIGHT do
363 */
364 unsigned int cur_depth;
365
366 /*
367 * io_u's about to be committed
368 */
369 unsigned int io_u_queued;
370
371 /*
372 * io_u's submitted but not completed yet
373 */
374 unsigned int io_u_in_flight;
375
376 /*
377 * List of free and busy io_u's
378 */
379 struct flist_head io_u_freelist;
380 struct flist_head io_u_busylist;
381 struct flist_head io_u_requeues;
382 pthread_mutex_t io_u_lock;
383 pthread_cond_t free_cond;
384
385 /*
386 * async verify offload
387 */
388 struct flist_head verify_list;
389 pthread_t *verify_threads;
390 unsigned int nr_verify_threads;
391 pthread_cond_t verify_cond;
392 int verify_thread_exit;
393
394 /*
395 * Rate state
396 */
397 unsigned long long rate_bps[2];
398 long rate_pending_usleep[2];
399 unsigned long rate_bytes[2];
400 unsigned long rate_blocks[2];
401 struct timeval lastrate[2];
402
403 unsigned long long total_io_size;
404 unsigned long long fill_device_size;
405
406 unsigned long io_issues[2];
407 unsigned long long io_blocks[2];
408 unsigned long long this_io_blocks[2];
409 unsigned long long io_bytes[2];
410 unsigned long long io_skip_bytes;
411 unsigned long long this_io_bytes[2];
412 unsigned long long zone_bytes;
413 struct fio_mutex *mutex;
414
415 /*
416 * State for random io, a bitmap of blocks done vs not done
417 */
418 union {
419 os_random_state_t random_state;
420 struct frand_state __random_state;
421 };
422
423 struct timeval start; /* start of this loop */
424 struct timeval epoch; /* time job was started */
425 struct timeval last_issue;
426 struct timeval tv_cache;
427 unsigned int tv_cache_nr;
428 unsigned int tv_cache_mask;
429 unsigned int ramp_time_over;
430
431 /*
432 * read/write mixed workload state
433 */
434 union {
435 os_random_state_t rwmix_state;
436 struct frand_state __rwmix_state;
437 };
438 unsigned long rwmix_issues;
439 enum fio_ddir rwmix_ddir;
440 unsigned int ddir_seq_nr;
441
442 /*
443 * IO history logs for verification. We use a tree for sorting,
444 * if we are overwriting. Otherwise just use a fifo.
445 */
446 struct rb_root io_hist_tree;
447 struct flist_head io_hist_list;
448 unsigned long io_hist_len;
449
450 /*
451 * For IO replaying
452 */
453 struct flist_head io_log_list;
454
455 /*
456 * For tracking/handling discards
457 */
458 struct flist_head trim_list;
459 unsigned long trim_entries;
460
461 /*
462 * for fileservice, how often to switch to a new file
463 */
464 unsigned int file_service_nr;
465 unsigned int file_service_left;
466 struct fio_file *file_service_file;
467
468 unsigned int sync_file_range_nr;
469
470 /*
471 * For generating file sizes
472 */
473 union {
474 os_random_state_t file_size_state;
475 struct frand_state __file_size_state;
476 };
477
478 /*
479 * Error counts
480 */
481 unsigned int total_err_count;
482 int first_error;
483
484 struct fio_flow *flow;
485
486 /*
487 * Can be overloaded by profiles
488 */
489 struct prof_io_ops prof_io_ops;
490 void *prof_data;
491};
492
493/*
494 * when should interactive ETA output be generated
495 */
496enum {
497 FIO_ETA_AUTO,
498 FIO_ETA_ALWAYS,
499 FIO_ETA_NEVER,
500};
501
502#define __td_verror(td, err, msg, func) \
503 do { \
504 if ((td)->error) \
505 break; \
506 int e = (err); \
507 (td)->error = e; \
508 if (!(td)->first_error) \
509 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
510 } while (0)
511
512
513#define td_clear_error(td) \
514 (td)->error = 0;
515#define td_verror(td, err, func) \
516 __td_verror((td), (err), strerror((err)), (func))
517#define td_vmsg(td, err, msg, func) \
518 __td_verror((td), (err), (msg), (func))
519
520#define __fio_stringify_1(x) #x
521#define __fio_stringify(x) __fio_stringify_1(x)
522
523extern int exitall_on_terminate;
524extern unsigned int thread_number;
525extern unsigned int nr_process, nr_thread;
526extern int shm_id;
527extern int groupid;
528extern int terse_output;
529extern int temp_stall_ts;
530extern unsigned long long mlock_size;
531extern uintptr_t page_mask, page_size;
532extern int read_only;
533extern int eta_print;
534extern unsigned long done_secs;
535extern char *job_section;
536extern int fio_gtod_offload;
537extern int fio_gtod_cpu;
538extern enum fio_cs fio_clock_source;
539extern int warnings_fatal;
540extern int terse_version;
541extern int is_backend;
542extern int nr_clients;
543extern int log_syslog;
544extern const char fio_version_string[];
545extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
546
547extern struct thread_data *threads;
548
549static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
550{
551 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
552}
553
554#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
555#define TO_MAP_BLOCK(f, b) (b)
556#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
557#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
558
559#define REAL_MAX_JOBS 2048
560
561#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
562
563static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
564{
565 if (err == EILSEQ)
566 return ERROR_TYPE_VERIFY;
567 if (ddir == DDIR_READ)
568 return ERROR_TYPE_READ;
569 return ERROR_TYPE_WRITE;
570}
571
572static inline void update_error_count(struct thread_data *td, int err)
573{
574 td->total_err_count++;
575 if (td->total_err_count == 1)
576 td->first_error = err;
577}
578
579static inline int should_fsync(struct thread_data *td)
580{
581 if (td->last_was_sync)
582 return 0;
583 if (td->o.odirect)
584 return 0;
585 if (td_write(td) || td_rw(td) || td->o.override_sync)
586 return 1;
587
588 return 0;
589}
590
591/*
592 * Init/option functions
593 */
594extern int __must_check parse_options(int, char **);
595extern int parse_jobs_ini(char *, int, int);
596extern int parse_cmd_line(int, char **);
597extern int fio_backend(void);
598extern void reset_fio_state(void);
599extern void clear_io_state(struct thread_data *);
600extern int fio_options_parse(struct thread_data *, char **, int);
601extern void fio_keywords_init(void);
602extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
603extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
604extern void fio_fill_default_options(struct thread_data *);
605extern int fio_show_option_help(const char *);
606extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
607extern void fio_options_dup_and_init(struct option *);
608extern void fio_options_mem_dupe(struct thread_data *);
609extern void options_mem_dupe(void *data, struct fio_option *options);
610extern void td_fill_rand_seeds(struct thread_data *);
611extern void add_job_opts(const char **);
612extern char *num2str(unsigned long, int, int, int);
613extern int ioengine_load(struct thread_data *);
614
615#define FIO_GETOPT_JOB 0x89000000
616#define FIO_GETOPT_IOENGINE 0x98000000
617#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
618
619/*
620 * ETA/status stuff
621 */
622extern void print_thread_status(void);
623extern void print_status_init(int);
624
625/*
626 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
627 * will never back again. It may cycle between running/verififying/fsyncing.
628 * Once the thread reaches TD_EXITED, it is just waiting for the core to
629 * reap it.
630 */
631enum {
632 TD_NOT_CREATED = 0,
633 TD_CREATED,
634 TD_INITIALIZED,
635 TD_RAMP,
636 TD_RUNNING,
637 TD_PRE_READING,
638 TD_VERIFYING,
639 TD_FSYNCING,
640 TD_EXITED,
641 TD_REAPED,
642};
643
644extern void td_set_runstate(struct thread_data *, int);
645#define TERMINATE_ALL (-1)
646extern void fio_terminate_threads(int);
647
648/*
649 * Memory helpers
650 */
651extern int __must_check fio_pin_memory(void);
652extern void fio_unpin_memory(void);
653extern int __must_check allocate_io_mem(struct thread_data *);
654extern void free_io_mem(struct thread_data *);
655
656/*
657 * Reset stats after ramp time completes
658 */
659extern void reset_all_stats(struct thread_data *);
660
661/*
662 * blktrace support
663 */
664#ifdef FIO_HAVE_BLKTRACE
665extern int is_blktrace(const char *);
666extern int load_blktrace(struct thread_data *, const char *);
667#endif
668
669/*
670 * Mark unused variables passed to ops functions as unused, to silence gcc
671 */
672#define fio_unused __attribute((__unused__))
673#define fio_init __attribute__((constructor))
674#define fio_exit __attribute__((destructor))
675
676#define for_each_td(td, i) \
677 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
678#define for_each_file(td, f, i) \
679 if ((td)->files_index) \
680 for ((i) = 0, (f) = (td)->files[0]; \
681 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
682 (i)++)
683
684#define fio_assert(td, cond) do { \
685 if (!(cond)) { \
686 int *__foo = NULL; \
687 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
688 td_set_runstate((td), TD_EXITED); \
689 (td)->error = EFAULT; \
690 *__foo = 0; \
691 } \
692} while (0)
693
694static inline int fio_fill_issue_time(struct thread_data *td)
695{
696 if (td->o.read_iolog_file ||
697 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
698 return 1;
699
700 return 0;
701}
702
703static inline int __should_check_rate(struct thread_data *td,
704 enum fio_ddir ddir)
705{
706 struct thread_options *o = &td->o;
707
708 /*
709 * If some rate setting was given, we need to check it
710 */
711 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
712 o->rate_iops_min[ddir])
713 return 1;
714
715 return 0;
716}
717
718static inline int should_check_rate(struct thread_data *td,
719 unsigned long *bytes_done)
720{
721 int ret = 0;
722
723 if (bytes_done[0])
724 ret |= __should_check_rate(td, 0);
725 if (bytes_done[1])
726 ret |= __should_check_rate(td, 1);
727
728 return ret;
729}
730
731static inline int is_power_of_2(unsigned int val)
732{
733 return (val != 0 && ((val & (val - 1)) == 0));
734}
735
736/*
737 * We currently only need to do locking if we have verifier threads
738 * accessing our internal structures too
739 */
740static inline void td_io_u_lock(struct thread_data *td)
741{
742 if (td->o.verify_async)
743 pthread_mutex_lock(&td->io_u_lock);
744}
745
746static inline void td_io_u_unlock(struct thread_data *td)
747{
748 if (td->o.verify_async)
749 pthread_mutex_unlock(&td->io_u_lock);
750}
751
752static inline void td_io_u_free_notify(struct thread_data *td)
753{
754 if (td->o.verify_async)
755 pthread_cond_signal(&td->free_cond);
756}
757
758extern const char *fio_get_arch_string(int);
759extern const char *fio_get_os_string(int);
760
761#endif