Add option to skip delays when replaying traces
[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
17#include "compiler/compiler.h"
18#include "flist.h"
19#include "fifo.h"
20#include "rbtree.h"
21#include "arch/arch.h"
22#include "os/os.h"
23#include "mutex.h"
24#include "log.h"
25#include "debug.h"
26#include "file.h"
27#include "io_ddir.h"
28#include "ioengine.h"
29#include "iolog.h"
30#include "helpers.h"
31#include "options.h"
32#include "profile.h"
33#include "time.h"
34#include "lib/getopt.h"
35
36#ifdef FIO_HAVE_GUASI
37#include <guasi.h>
38#endif
39
40#ifdef FIO_HAVE_SOLARISAIO
41#include <sys/asynch.h>
42#endif
43
44struct group_run_stats {
45 unsigned long long max_run[2], min_run[2];
46 unsigned long long max_bw[2], min_bw[2];
47 unsigned long long io_kb[2];
48 unsigned long long agg[2];
49 unsigned int kb_base;
50};
51
52/*
53 * What type of allocation to use for io buffers
54 */
55enum fio_memtype {
56 MEM_MALLOC = 0, /* ordinary malloc */
57 MEM_SHM, /* use shared memory segments */
58 MEM_SHMHUGE, /* use shared memory segments with huge pages */
59 MEM_MMAP, /* use anonynomous mmap */
60 MEM_MMAPHUGE, /* memory mapped huge file */
61};
62
63/*
64 * offset generator types
65 */
66enum {
67 RW_SEQ_SEQ = 0,
68 RW_SEQ_IDENT,
69};
70
71/*
72 * How many depth levels to log
73 */
74#define FIO_IO_U_MAP_NR 8
75#define FIO_IO_U_LAT_U_NR 10
76#define FIO_IO_U_LAT_M_NR 12
77
78#define MAX_PATTERN_SIZE 512
79
80struct thread_stat {
81 char *name;
82 char *verror;
83 int error;
84 int groupid;
85 pid_t pid;
86 char *description;
87 int members;
88
89 struct io_log *slat_log;
90 struct io_log *clat_log;
91 struct io_log *lat_log;
92 struct io_log *bw_log;
93
94 /*
95 * bandwidth and latency stats
96 */
97 struct io_stat clat_stat[2]; /* completion latency */
98 struct io_stat slat_stat[2]; /* submission latency */
99 struct io_stat lat_stat[2]; /* total latency */
100 struct io_stat bw_stat[2]; /* bandwidth stats */
101
102 unsigned long long stat_io_bytes[2];
103 struct timeval stat_sample_time[2];
104
105 /*
106 * fio system usage accounting
107 */
108 struct rusage ru_start;
109 struct rusage ru_end;
110 unsigned long usr_time;
111 unsigned long sys_time;
112 unsigned long ctx;
113 unsigned long minf, majf;
114
115 /*
116 * IO depth and latency stats
117 */
118 unsigned int io_u_map[FIO_IO_U_MAP_NR];
119 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
120 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
121 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
122 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
123 unsigned long total_io_u[2];
124 unsigned long short_io_u[2];
125 unsigned long total_submit;
126 unsigned long total_complete;
127
128 unsigned long long io_bytes[2];
129 unsigned long runtime[2];
130 unsigned long total_run_time;
131
132 /*
133 * IO Error related stats
134 */
135 unsigned continue_on_error;
136 unsigned long total_err_count;
137 int first_error;
138
139 unsigned int kb_base;
140};
141
142struct bssplit {
143 unsigned int bs;
144 unsigned char perc;
145};
146
147struct thread_options {
148 int pad;
149 char *description;
150 char *name;
151 char *directory;
152 char *filename;
153 char *opendir;
154 char *ioengine;
155 enum td_ddir td_ddir;
156 unsigned int rw_seq;
157 unsigned int kb_base;
158 unsigned int ddir_seq_nr;
159 unsigned int iodepth;
160 unsigned int iodepth_low;
161 unsigned int iodepth_batch;
162 unsigned int iodepth_batch_complete;
163
164 unsigned long long size;
165 unsigned int fill_device;
166 unsigned long long file_size_low;
167 unsigned long long file_size_high;
168 unsigned long long start_offset;
169
170 unsigned int bs[2];
171 unsigned int ba[2];
172 unsigned int min_bs[2];
173 unsigned int max_bs[2];
174 struct bssplit *bssplit[2];
175 unsigned int bssplit_nr[2];
176
177 unsigned int nr_files;
178 unsigned int open_files;
179 enum file_lock_mode file_lock_mode;
180 unsigned int lockfile_batch;
181
182 unsigned int odirect;
183 unsigned int invalidate_cache;
184 unsigned int create_serialize;
185 unsigned int create_fsync;
186 unsigned int create_on_open;
187 unsigned int end_fsync;
188 unsigned int pre_read;
189 unsigned int sync_io;
190 unsigned int verify;
191 unsigned int do_verify;
192 unsigned int verifysort;
193 unsigned int verify_interval;
194 unsigned int verify_offset;
195 char verify_pattern[MAX_PATTERN_SIZE];
196 unsigned int verify_pattern_bytes;
197 unsigned int verify_fatal;
198 unsigned int verify_async;
199 unsigned long long verify_backlog;
200 unsigned int verify_batch;
201 unsigned int use_thread;
202 unsigned int unlink;
203 unsigned int do_disk_util;
204 unsigned int override_sync;
205 unsigned int rand_repeatable;
206 unsigned int write_lat_log;
207 unsigned int write_bw_log;
208 unsigned int norandommap;
209 unsigned int softrandommap;
210 unsigned int bs_unaligned;
211 unsigned int fsync_on_close;
212
213 unsigned int hugepage_size;
214 unsigned int rw_min_bs;
215 unsigned int thinktime;
216 unsigned int thinktime_spin;
217 unsigned int thinktime_blocks;
218 unsigned int fsync_blocks;
219 unsigned int fdatasync_blocks;
220 unsigned long start_delay;
221 unsigned long long timeout;
222 unsigned long long ramp_time;
223 unsigned int overwrite;
224 unsigned int bw_avg_time;
225 unsigned int loops;
226 unsigned long long zone_size;
227 unsigned long long zone_skip;
228 enum fio_memtype mem_type;
229 unsigned int mem_align;
230
231 unsigned int stonewall;
232 unsigned int new_group;
233 unsigned int numjobs;
234 os_cpu_mask_t cpumask;
235 unsigned int cpumask_set;
236 os_cpu_mask_t verify_cpumask;
237 unsigned int verify_cpumask_set;
238 unsigned int iolog;
239 unsigned int rwmixcycle;
240 unsigned int rwmix[2];
241 unsigned int nice;
242 unsigned int file_service_type;
243 unsigned int group_reporting;
244 unsigned int fadvise_hint;
245 unsigned int fallocate;
246 unsigned int zero_buffers;
247 unsigned int refill_buffers;
248 unsigned int time_based;
249 unsigned int disable_lat;
250 unsigned int disable_clat;
251 unsigned int disable_slat;
252 unsigned int disable_bw;
253 unsigned int gtod_reduce;
254 unsigned int gtod_cpu;
255 unsigned int gtod_offload;
256 enum fio_cs clocksource;
257 unsigned int no_stall;
258
259 char *read_iolog_file;
260 char *write_iolog_file;
261 char *bw_log_file;
262 char *lat_log_file;
263
264 /*
265 * Pre-run and post-run shell
266 */
267 char *exec_prerun;
268 char *exec_postrun;
269
270 unsigned int rate[2];
271 unsigned int ratemin[2];
272 unsigned int ratecycle;
273 unsigned int rate_iops[2];
274 unsigned int rate_iops_min[2];
275
276 char *ioscheduler;
277
278 /*
279 * CPU "io" cycle burner
280 */
281 unsigned int cpuload;
282 unsigned int cpucycle;
283
284 /*
285 * I/O Error handling
286 */
287 unsigned int continue_on_error;
288
289 /*
290 * Benchmark profile type
291 */
292 char *profile;
293
294 /*
295 * blkio cgroup support
296 */
297 char *cgroup;
298 unsigned int cgroup_weight;
299 unsigned int cgroup_nodelete;
300
301 unsigned int uid;
302 unsigned int gid;
303
304 unsigned int sync_file_range;
305};
306
307#define FIO_VERROR_SIZE 128
308
309/*
310 * This describes a single thread/process executing a fio job.
311 */
312struct thread_data {
313 struct thread_options o;
314 char verror[FIO_VERROR_SIZE];
315 pthread_t thread;
316 int thread_number;
317 int groupid;
318 struct thread_stat ts;
319 struct fio_file **files;
320 unsigned int files_size;
321 unsigned int files_index;
322 unsigned int nr_open_files;
323 unsigned int nr_done_files;
324 unsigned int nr_normal_files;
325 union {
326 unsigned int next_file;
327 os_random_state_t next_file_state;
328 };
329 int error;
330 int done;
331 pid_t pid;
332 char *orig_buffer;
333 size_t orig_buffer_size;
334 volatile int terminate;
335 volatile int runstate;
336 unsigned int ioprio;
337 unsigned int ioprio_set;
338 unsigned int last_was_sync;
339 enum fio_ddir last_ddir;
340
341 char *mmapfile;
342 int mmapfd;
343
344 void *iolog_buf;
345 FILE *iolog_f;
346
347 char *sysfs_root;
348
349 unsigned long rand_seeds[6];
350
351 os_random_state_t bsrange_state;
352 os_random_state_t verify_state;
353
354 unsigned int verify_batch;
355
356 int shm_id;
357
358 /*
359 * IO engine hooks, contains everything needed to submit an io_u
360 * to any of the available IO engines.
361 */
362 struct ioengine_ops *io_ops;
363
364 /*
365 * Current IO depth and list of free and busy io_u's.
366 */
367 unsigned int cur_depth;
368 unsigned int io_u_queued;
369 struct flist_head io_u_freelist;
370 struct flist_head io_u_busylist;
371 struct flist_head io_u_requeues;
372 pthread_mutex_t io_u_lock;
373 pthread_cond_t free_cond;
374
375 /*
376 * async verify offload
377 */
378 struct flist_head verify_list;
379 pthread_t *verify_threads;
380 unsigned int nr_verify_threads;
381 pthread_cond_t verify_cond;
382 int verify_thread_exit;
383
384 /*
385 * Rate state
386 */
387 unsigned long rate_nsec_cycle[2];
388 long rate_pending_usleep[2];
389 unsigned long rate_bytes[2];
390 unsigned long rate_blocks[2];
391 struct timeval lastrate[2];
392
393 unsigned long long total_io_size;
394 unsigned long long fill_device_size;
395
396 unsigned long io_issues[2];
397 unsigned long long io_blocks[2];
398 unsigned long long io_bytes[2];
399 unsigned long long io_skip_bytes;
400 unsigned long long this_io_bytes[2];
401 unsigned long long zone_bytes;
402 struct fio_mutex *mutex;
403
404 /*
405 * State for random io, a bitmap of blocks done vs not done
406 */
407 os_random_state_t random_state;
408
409 struct timeval start; /* start of this loop */
410 struct timeval epoch; /* time job was started */
411 struct timeval last_issue;
412 struct timeval tv_cache;
413 unsigned int tv_cache_nr;
414 unsigned int tv_cache_mask;
415 unsigned int ramp_time_over;
416
417 /*
418 * read/write mixed workload state
419 */
420 os_random_state_t rwmix_state;
421 unsigned long rwmix_issues;
422 enum fio_ddir rwmix_ddir;
423 unsigned int ddir_seq_nr;
424
425 /*
426 * IO history logs for verification. We use a tree for sorting,
427 * if we are overwriting. Otherwise just use a fifo.
428 */
429 struct rb_root io_hist_tree;
430 struct flist_head io_hist_list;
431 unsigned long io_hist_len;
432
433 /*
434 * For IO replaying
435 */
436 struct flist_head io_log_list;
437
438 /*
439 * for fileservice, how often to switch to a new file
440 */
441 unsigned int file_service_nr;
442 unsigned int file_service_left;
443 struct fio_file *file_service_file;
444
445 unsigned int sync_file_range_nr;
446
447 /*
448 * For generating file sizes
449 */
450 os_random_state_t file_size_state;
451
452 /*
453 * Error counts
454 */
455 unsigned int total_err_count;
456 int first_error;
457
458 /*
459 * Can be overloaded by profiles
460 */
461 struct prof_io_ops prof_io_ops;
462 void *prof_data;
463};
464
465/*
466 * when should interactive ETA output be generated
467 */
468enum {
469 FIO_ETA_AUTO,
470 FIO_ETA_ALWAYS,
471 FIO_ETA_NEVER,
472};
473
474#define __td_verror(td, err, msg, func) \
475 do { \
476 if ((td)->error) \
477 break; \
478 int e = (err); \
479 (td)->error = e; \
480 if (!(td)->first_error) \
481 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
482 } while (0)
483
484
485#define td_clear_error(td) \
486 (td)->error = 0;
487#define td_verror(td, err, func) \
488 __td_verror((td), (err), strerror((err)), (func))
489#define td_vmsg(td, err, msg, func) \
490 __td_verror((td), (err), (msg), (func))
491
492extern int exitall_on_terminate;
493extern int thread_number;
494extern int nr_process, nr_thread;
495extern int shm_id;
496extern int groupid;
497extern int terse_output;
498extern int temp_stall_ts;
499extern unsigned long long mlock_size;
500extern unsigned long page_mask, page_size;
501extern int read_only;
502extern int eta_print;
503extern unsigned long done_secs;
504extern char *job_section;
505extern int fio_gtod_offload;
506extern int fio_gtod_cpu;
507extern enum fio_cs fio_clock_source;
508
509extern struct thread_data *threads;
510
511static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
512{
513 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
514}
515
516#define BLOCKS_PER_MAP (8 * sizeof(int))
517#define TO_MAP_BLOCK(f, b) (b)
518#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
519#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
520
521#define MAX_JOBS (1024)
522
523#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
524
525static inline void update_error_count(struct thread_data *td, int err)
526{
527 td->total_err_count++;
528 if (td->total_err_count == 1)
529 td->first_error = err;
530}
531
532static inline int should_fsync(struct thread_data *td)
533{
534 if (td->last_was_sync)
535 return 0;
536 if (td->o.odirect)
537 return 0;
538 if (td_write(td) || td_rw(td) || td->o.override_sync)
539 return 1;
540
541 return 0;
542}
543
544/*
545 * Init/option functions
546 */
547extern int __must_check parse_options(int, char **);
548extern int fio_options_parse(struct thread_data *, char **, int);
549extern void fio_keywords_init(void);
550extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
551extern void fio_fill_default_options(struct thread_data *);
552extern int fio_show_option_help(const char *);
553extern void fio_options_dup_and_init(struct option *);
554extern void options_mem_dupe(struct thread_data *);
555extern void options_mem_free(struct thread_data *);
556extern void td_fill_rand_seeds(struct thread_data *);
557extern void add_job_opts(const char **);
558#define FIO_GETOPT_JOB 0x89988998
559#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
560
561/*
562 * ETA/status stuff
563 */
564extern void print_thread_status(void);
565extern void print_status_init(int);
566
567/*
568 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
569 * will never back again. It may cycle between running/verififying/fsyncing.
570 * Once the thread reaches TD_EXITED, it is just waiting for the core to
571 * reap it.
572 */
573enum {
574 TD_NOT_CREATED = 0,
575 TD_CREATED,
576 TD_INITIALIZED,
577 TD_RAMP,
578 TD_RUNNING,
579 TD_PRE_READING,
580 TD_VERIFYING,
581 TD_FSYNCING,
582 TD_EXITED,
583 TD_REAPED,
584};
585
586extern void td_set_runstate(struct thread_data *, int);
587
588/*
589 * Memory helpers
590 */
591extern int __must_check fio_pin_memory(void);
592extern void fio_unpin_memory(void);
593extern int __must_check allocate_io_mem(struct thread_data *);
594extern void free_io_mem(struct thread_data *);
595
596/*
597 * Reset stats after ramp time completes
598 */
599extern void reset_all_stats(struct thread_data *);
600
601/*
602 * blktrace support
603 */
604#ifdef FIO_HAVE_BLKTRACE
605extern int is_blktrace(const char *);
606extern int load_blktrace(struct thread_data *, const char *);
607#endif
608
609/*
610 * Mark unused variables passed to ops functions as unused, to silence gcc
611 */
612#define fio_unused __attribute((__unused__))
613#define fio_init __attribute__((constructor))
614#define fio_exit __attribute__((destructor))
615
616#define for_each_td(td, i) \
617 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
618#define for_each_file(td, f, i) \
619 if ((td)->files_index) \
620 for ((i) = 0, (f) = (td)->files[0]; \
621 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
622 (i)++)
623
624#define fio_assert(td, cond) do { \
625 if (!(cond)) { \
626 int *__foo = NULL; \
627 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
628 (td)->runstate = TD_EXITED; \
629 (td)->error = EFAULT; \
630 *__foo = 0; \
631 } \
632} while (0)
633
634static inline int fio_fill_issue_time(struct thread_data *td)
635{
636 if (td->o.read_iolog_file ||
637 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
638 return 1;
639
640 return 0;
641}
642
643/*
644 * Cheesy number->string conversion, complete with carry rounding error.
645 */
646static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
647{
648 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
649 unsigned int thousand;
650 char *buf;
651 int i;
652
653 if (pow2)
654 thousand = 1024;
655 else
656 thousand = 1000;
657
658 buf = malloc(128);
659
660 for (i = 0; base > 1; i++)
661 base /= thousand;
662
663 do {
664 int len, carry = 0;
665
666 len = sprintf(buf, "%'lu", num);
667 if (len <= maxlen) {
668 if (i >= 1) {
669 buf[len] = postfix[i];
670 buf[len + 1] = '\0';
671 }
672 return buf;
673 }
674
675 if ((num % thousand) >= (thousand / 2))
676 carry = 1;
677
678 num /= thousand;
679 num += carry;
680 i++;
681 } while (i <= 5);
682
683 return buf;
684}
685
686static inline int __should_check_rate(struct thread_data *td,
687 enum fio_ddir ddir)
688{
689 struct thread_options *o = &td->o;
690
691 /*
692 * If some rate setting was given, we need to check it
693 */
694 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
695 o->rate_iops_min[ddir])
696 return 1;
697
698 return 0;
699}
700
701static inline int should_check_rate(struct thread_data *td,
702 unsigned long *bytes_done)
703{
704 int ret = 0;
705
706 if (bytes_done[0])
707 ret |= __should_check_rate(td, 0);
708 if (bytes_done[1])
709 ret |= __should_check_rate(td, 1);
710
711 return ret;
712}
713
714static inline int is_power_of_2(unsigned int val)
715{
716 return (val != 0 && ((val & (val - 1)) == 0));
717}
718
719/*
720 * We currently only need to do locking if we have verifier threads
721 * accessing our internal structures too
722 */
723static inline void td_io_u_lock(struct thread_data *td)
724{
725 if (td->o.verify_async)
726 pthread_mutex_lock(&td->io_u_lock);
727}
728
729static inline void td_io_u_unlock(struct thread_data *td)
730{
731 if (td->o.verify_async)
732 pthread_mutex_unlock(&td->io_u_lock);
733}
734
735static inline void td_io_u_free_notify(struct thread_data *td)
736{
737 if (td->o.verify_async)
738 pthread_cond_signal(&td->free_cond);
739}
740
741#endif