Improve accuracy of rate= option
[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#include "server.h"
39#include "stat.h"
40
41#ifdef FIO_HAVE_GUASI
42#include <guasi.h>
43#endif
44
45#ifdef FIO_HAVE_SOLARISAIO
46#include <sys/asynch.h>
47#endif
48
49/*
50 * What type of allocation to use for io buffers
51 */
52enum fio_memtype {
53 MEM_MALLOC = 0, /* ordinary malloc */
54 MEM_SHM, /* use shared memory segments */
55 MEM_SHMHUGE, /* use shared memory segments with huge pages */
56 MEM_MMAP, /* use anonynomous mmap */
57 MEM_MMAPHUGE, /* memory mapped huge file */
58};
59
60/*
61 * offset generator types
62 */
63enum {
64 RW_SEQ_SEQ = 0,
65 RW_SEQ_IDENT,
66};
67
68struct bssplit {
69 unsigned int bs;
70 unsigned char perc;
71};
72
73struct thread_options {
74 int pad;
75 char *description;
76 char *name;
77 char *directory;
78 char *filename;
79 char *opendir;
80 char *ioengine;
81 enum td_ddir td_ddir;
82 unsigned int rw_seq;
83 unsigned int kb_base;
84 unsigned int ddir_seq_nr;
85 long ddir_seq_add;
86 unsigned int iodepth;
87 unsigned int iodepth_low;
88 unsigned int iodepth_batch;
89 unsigned int iodepth_batch_complete;
90
91 unsigned long long size;
92 unsigned int size_percent;
93 unsigned int fill_device;
94 unsigned long long file_size_low;
95 unsigned long long file_size_high;
96 unsigned long long start_offset;
97
98 unsigned int bs[2];
99 unsigned int ba[2];
100 unsigned int min_bs[2];
101 unsigned int max_bs[2];
102 struct bssplit *bssplit[2];
103 unsigned int bssplit_nr[2];
104
105 unsigned int nr_files;
106 unsigned int open_files;
107 enum file_lock_mode file_lock_mode;
108 unsigned int lockfile_batch;
109
110 unsigned int odirect;
111 unsigned int invalidate_cache;
112 unsigned int create_serialize;
113 unsigned int create_fsync;
114 unsigned int create_on_open;
115 unsigned int end_fsync;
116 unsigned int pre_read;
117 unsigned int sync_io;
118 unsigned int verify;
119 unsigned int do_verify;
120 unsigned int verifysort;
121 unsigned int verify_interval;
122 unsigned int verify_offset;
123 char verify_pattern[MAX_PATTERN_SIZE];
124 unsigned int verify_pattern_bytes;
125 unsigned int verify_fatal;
126 unsigned int verify_dump;
127 unsigned int verify_async;
128 unsigned long long verify_backlog;
129 unsigned int verify_batch;
130 unsigned int use_thread;
131 unsigned int unlink;
132 unsigned int do_disk_util;
133 unsigned int override_sync;
134 unsigned int rand_repeatable;
135 unsigned int use_os_rand;
136 unsigned int write_lat_log;
137 unsigned int write_bw_log;
138 unsigned int write_iops_log;
139 unsigned int norandommap;
140 unsigned int softrandommap;
141 unsigned int bs_unaligned;
142 unsigned int fsync_on_close;
143
144 unsigned int hugepage_size;
145 unsigned int rw_min_bs;
146 unsigned int thinktime;
147 unsigned int thinktime_spin;
148 unsigned int thinktime_blocks;
149 unsigned int fsync_blocks;
150 unsigned int fdatasync_blocks;
151 unsigned int barrier_blocks;
152 unsigned long long start_delay;
153 unsigned long long timeout;
154 unsigned long long ramp_time;
155 unsigned int overwrite;
156 unsigned int bw_avg_time;
157 unsigned int iops_avg_time;
158 unsigned int loops;
159 unsigned long long zone_size;
160 unsigned long long zone_skip;
161 enum fio_memtype mem_type;
162 unsigned int mem_align;
163
164 unsigned int stonewall;
165 unsigned int new_group;
166 unsigned int numjobs;
167 os_cpu_mask_t cpumask;
168 unsigned int cpumask_set;
169 os_cpu_mask_t verify_cpumask;
170 unsigned int verify_cpumask_set;
171 unsigned int iolog;
172 unsigned int rwmixcycle;
173 unsigned int rwmix[2];
174 unsigned int nice;
175 unsigned int file_service_type;
176 unsigned int group_reporting;
177 unsigned int fadvise_hint;
178 enum fio_fallocate_mode fallocate_mode;
179 unsigned int zero_buffers;
180 unsigned int refill_buffers;
181 unsigned int scramble_buffers;
182 unsigned int time_based;
183 unsigned int disable_lat;
184 unsigned int disable_clat;
185 unsigned int disable_slat;
186 unsigned int disable_bw;
187 unsigned int gtod_reduce;
188 unsigned int gtod_cpu;
189 unsigned int gtod_offload;
190 enum fio_cs clocksource;
191 unsigned int no_stall;
192 unsigned int trim_percentage;
193 unsigned int trim_batch;
194 unsigned int trim_zero;
195 unsigned long long trim_backlog;
196 unsigned int clat_percentiles;
197 unsigned int overwrite_plist;
198 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
199
200 char *read_iolog_file;
201 char *write_iolog_file;
202 char *bw_log_file;
203 char *lat_log_file;
204 char *iops_log_file;
205 char *replay_redirect;
206
207 /*
208 * Pre-run and post-run shell
209 */
210 char *exec_prerun;
211 char *exec_postrun;
212
213 unsigned int rate[2];
214 unsigned int ratemin[2];
215 unsigned int ratecycle;
216 unsigned int rate_iops[2];
217 unsigned int rate_iops_min[2];
218
219 char *ioscheduler;
220
221 /*
222 * CPU "io" cycle burner
223 */
224 unsigned int cpuload;
225 unsigned int cpucycle;
226
227 /*
228 * I/O Error handling
229 */
230 unsigned int continue_on_error;
231
232 /*
233 * Benchmark profile type
234 */
235 char *profile;
236
237 /*
238 * blkio cgroup support
239 */
240 char *cgroup;
241 unsigned int cgroup_weight;
242 unsigned int cgroup_nodelete;
243
244 unsigned int uid;
245 unsigned int gid;
246
247 unsigned int sync_file_range;
248
249 unsigned int userspace_libaio_reap;
250};
251
252/*
253 * This describes a single thread/process executing a fio job.
254 */
255struct thread_data {
256 struct thread_options o;
257 char verror[FIO_VERROR_SIZE];
258 pthread_t thread;
259 int thread_number;
260 int groupid;
261 struct thread_stat ts;
262
263 struct io_log *slat_log;
264 struct io_log *clat_log;
265 struct io_log *lat_log;
266 struct io_log *bw_log;
267 struct io_log *iops_log;
268
269 uint64_t stat_io_bytes[2];
270 struct timeval bw_sample_time;
271
272 uint64_t stat_io_blocks[2];
273 struct timeval iops_sample_time;
274
275 struct rusage ru_start;
276 struct rusage ru_end;
277
278 struct fio_file **files;
279 unsigned int files_size;
280 unsigned int files_index;
281 unsigned int nr_open_files;
282 unsigned int nr_done_files;
283 unsigned int nr_normal_files;
284 union {
285 unsigned int next_file;
286 os_random_state_t next_file_state;
287 struct frand_state __next_file_state;
288 };
289 int error;
290 int done;
291 pid_t pid;
292 char *orig_buffer;
293 size_t orig_buffer_size;
294 volatile int terminate;
295 volatile int runstate;
296 unsigned int ioprio;
297 unsigned int ioprio_set;
298 unsigned int last_was_sync;
299 enum fio_ddir last_ddir;
300
301 char *mmapfile;
302 int mmapfd;
303
304 void *iolog_buf;
305 FILE *iolog_f;
306
307 char *sysfs_root;
308
309 unsigned long rand_seeds[8];
310
311 union {
312 os_random_state_t bsrange_state;
313 struct frand_state __bsrange_state;
314 };
315 union {
316 os_random_state_t verify_state;
317 struct frand_state __verify_state;
318 };
319 union {
320 os_random_state_t trim_state;
321 struct frand_state __trim_state;
322 };
323
324 struct frand_state buf_state;
325
326 unsigned int verify_batch;
327 unsigned int trim_batch;
328
329 int shm_id;
330
331 /*
332 * IO engine hooks, contains everything needed to submit an io_u
333 * to any of the available IO engines.
334 */
335 struct ioengine_ops *io_ops;
336
337 /*
338 * Current IO depth and list of free and busy io_u's.
339 */
340 unsigned int cur_depth;
341 unsigned int io_u_queued;
342 struct flist_head io_u_freelist;
343 struct flist_head io_u_busylist;
344 struct flist_head io_u_requeues;
345 pthread_mutex_t io_u_lock;
346 pthread_cond_t free_cond;
347
348 /*
349 * async verify offload
350 */
351 struct flist_head verify_list;
352 pthread_t *verify_threads;
353 unsigned int nr_verify_threads;
354 pthread_cond_t verify_cond;
355 int verify_thread_exit;
356
357 /*
358 * Rate state
359 */
360 unsigned long long rate_bps[2];
361 long rate_pending_usleep[2];
362 unsigned long rate_bytes[2];
363 unsigned long rate_blocks[2];
364 struct timeval lastrate[2];
365
366 unsigned long long total_io_size;
367 unsigned long long fill_device_size;
368
369 unsigned long io_issues[2];
370 unsigned long long io_blocks[2];
371 unsigned long long this_io_blocks[2];
372 unsigned long long io_bytes[2];
373 unsigned long long io_skip_bytes;
374 unsigned long long this_io_bytes[2];
375 unsigned long long zone_bytes;
376 struct fio_mutex *mutex;
377
378 /*
379 * State for random io, a bitmap of blocks done vs not done
380 */
381 union {
382 os_random_state_t random_state;
383 struct frand_state __random_state;
384 };
385
386 struct timeval start; /* start of this loop */
387 struct timeval epoch; /* time job was started */
388 struct timeval last_issue;
389 struct timeval tv_cache;
390 unsigned int tv_cache_nr;
391 unsigned int tv_cache_mask;
392 unsigned int ramp_time_over;
393
394 /*
395 * read/write mixed workload state
396 */
397 union {
398 os_random_state_t rwmix_state;
399 struct frand_state __rwmix_state;
400 };
401 unsigned long rwmix_issues;
402 enum fio_ddir rwmix_ddir;
403 unsigned int ddir_seq_nr;
404
405 /*
406 * IO history logs for verification. We use a tree for sorting,
407 * if we are overwriting. Otherwise just use a fifo.
408 */
409 struct rb_root io_hist_tree;
410 struct flist_head io_hist_list;
411 unsigned long io_hist_len;
412
413 /*
414 * For IO replaying
415 */
416 struct flist_head io_log_list;
417
418 /*
419 * For tracking/handling discards
420 */
421 struct flist_head trim_list;
422 unsigned long trim_entries;
423
424 /*
425 * for fileservice, how often to switch to a new file
426 */
427 unsigned int file_service_nr;
428 unsigned int file_service_left;
429 struct fio_file *file_service_file;
430
431 unsigned int sync_file_range_nr;
432
433 /*
434 * For generating file sizes
435 */
436 union {
437 os_random_state_t file_size_state;
438 struct frand_state __file_size_state;
439 };
440
441 /*
442 * Error counts
443 */
444 unsigned int total_err_count;
445 int first_error;
446
447 /*
448 * Can be overloaded by profiles
449 */
450 struct prof_io_ops prof_io_ops;
451 void *prof_data;
452};
453
454/*
455 * when should interactive ETA output be generated
456 */
457enum {
458 FIO_ETA_AUTO,
459 FIO_ETA_ALWAYS,
460 FIO_ETA_NEVER,
461};
462
463#define __td_verror(td, err, msg, func) \
464 do { \
465 if ((td)->error) \
466 break; \
467 int e = (err); \
468 (td)->error = e; \
469 if (!(td)->first_error) \
470 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
471 } while (0)
472
473
474#define td_clear_error(td) \
475 (td)->error = 0;
476#define td_verror(td, err, func) \
477 __td_verror((td), (err), strerror((err)), (func))
478#define td_vmsg(td, err, msg, func) \
479 __td_verror((td), (err), (msg), (func))
480
481#define __fio_stringify_1(x) #x
482#define __fio_stringify(x) __fio_stringify_1(x)
483
484extern int exitall_on_terminate;
485extern unsigned int thread_number;
486extern unsigned int nr_process, nr_thread;
487extern int shm_id;
488extern int groupid;
489extern int terse_output;
490extern int temp_stall_ts;
491extern unsigned long long mlock_size;
492extern unsigned long page_mask, page_size;
493extern int read_only;
494extern int eta_print;
495extern unsigned long done_secs;
496extern char *job_section;
497extern int fio_gtod_offload;
498extern int fio_gtod_cpu;
499extern enum fio_cs fio_clock_source;
500extern int warnings_fatal;
501extern int terse_version;
502extern int is_backend;
503extern int nr_clients;
504extern int log_syslog;
505extern const char fio_version_string[];
506extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
507
508extern struct thread_data *threads;
509
510static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
511{
512 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
513}
514
515#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
516#define TO_MAP_BLOCK(f, b) (b)
517#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
518#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
519
520#define REAL_MAX_JOBS 2048
521
522#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
523
524static inline void update_error_count(struct thread_data *td, int err)
525{
526 td->total_err_count++;
527 if (td->total_err_count == 1)
528 td->first_error = err;
529}
530
531static inline int should_fsync(struct thread_data *td)
532{
533 if (td->last_was_sync)
534 return 0;
535 if (td->o.odirect)
536 return 0;
537 if (td_write(td) || td_rw(td) || td->o.override_sync)
538 return 1;
539
540 return 0;
541}
542
543/*
544 * Init/option functions
545 */
546extern int __must_check parse_options(int, char **);
547extern int parse_jobs_ini(char *, int, int);
548extern int parse_cmd_line(int, char **);
549extern int exec_run(void);
550extern void reset_fio_state(void);
551extern int fio_options_parse(struct thread_data *, char **, int);
552extern void fio_keywords_init(void);
553extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
554extern void fio_fill_default_options(struct thread_data *);
555extern int fio_show_option_help(const char *);
556extern void fio_options_dup_and_init(struct option *);
557extern void options_mem_dupe(struct thread_data *);
558extern void options_mem_free(struct thread_data *);
559extern void td_fill_rand_seeds(struct thread_data *);
560extern void add_job_opts(const char **);
561extern char *num2str(unsigned long, int, int, int);
562
563#define FIO_GETOPT_JOB 0x89988998
564#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
565
566/*
567 * ETA/status stuff
568 */
569extern void print_thread_status(void);
570extern void print_status_init(int);
571
572/*
573 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
574 * will never back again. It may cycle between running/verififying/fsyncing.
575 * Once the thread reaches TD_EXITED, it is just waiting for the core to
576 * reap it.
577 */
578enum {
579 TD_NOT_CREATED = 0,
580 TD_CREATED,
581 TD_INITIALIZED,
582 TD_RAMP,
583 TD_RUNNING,
584 TD_PRE_READING,
585 TD_VERIFYING,
586 TD_FSYNCING,
587 TD_EXITED,
588 TD_REAPED,
589};
590
591extern void td_set_runstate(struct thread_data *, int);
592#define TERMINATE_ALL (-1)
593extern void fio_terminate_threads(int);
594
595/*
596 * Memory helpers
597 */
598extern int __must_check fio_pin_memory(void);
599extern void fio_unpin_memory(void);
600extern int __must_check allocate_io_mem(struct thread_data *);
601extern void free_io_mem(struct thread_data *);
602
603/*
604 * Reset stats after ramp time completes
605 */
606extern void reset_all_stats(struct thread_data *);
607
608/*
609 * blktrace support
610 */
611#ifdef FIO_HAVE_BLKTRACE
612extern int is_blktrace(const char *);
613extern int load_blktrace(struct thread_data *, const char *);
614#endif
615
616/*
617 * Mark unused variables passed to ops functions as unused, to silence gcc
618 */
619#define fio_unused __attribute((__unused__))
620#define fio_init __attribute__((constructor))
621#define fio_exit __attribute__((destructor))
622
623#define for_each_td(td, i) \
624 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
625#define for_each_file(td, f, i) \
626 if ((td)->files_index) \
627 for ((i) = 0, (f) = (td)->files[0]; \
628 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
629 (i)++)
630
631#define fio_assert(td, cond) do { \
632 if (!(cond)) { \
633 int *__foo = NULL; \
634 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
635 td_set_runstate((td), TD_EXITED); \
636 (td)->error = EFAULT; \
637 *__foo = 0; \
638 } \
639} while (0)
640
641static inline int fio_fill_issue_time(struct thread_data *td)
642{
643 if (td->o.read_iolog_file ||
644 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
645 return 1;
646
647 return 0;
648}
649
650static inline int __should_check_rate(struct thread_data *td,
651 enum fio_ddir ddir)
652{
653 struct thread_options *o = &td->o;
654
655 /*
656 * If some rate setting was given, we need to check it
657 */
658 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
659 o->rate_iops_min[ddir])
660 return 1;
661
662 return 0;
663}
664
665static inline int should_check_rate(struct thread_data *td,
666 unsigned long *bytes_done)
667{
668 int ret = 0;
669
670 if (bytes_done[0])
671 ret |= __should_check_rate(td, 0);
672 if (bytes_done[1])
673 ret |= __should_check_rate(td, 1);
674
675 return ret;
676}
677
678static inline int is_power_of_2(unsigned int val)
679{
680 return (val != 0 && ((val & (val - 1)) == 0));
681}
682
683/*
684 * We currently only need to do locking if we have verifier threads
685 * accessing our internal structures too
686 */
687static inline void td_io_u_lock(struct thread_data *td)
688{
689 if (td->o.verify_async)
690 pthread_mutex_lock(&td->io_u_lock);
691}
692
693static inline void td_io_u_unlock(struct thread_data *td)
694{
695 if (td->o.verify_async)
696 pthread_mutex_unlock(&td->io_u_lock);
697}
698
699static inline void td_io_u_free_notify(struct thread_data *td)
700{
701 if (td->o.verify_async)
702 pthread_cond_signal(&td->free_cond);
703}
704
705extern const char *fio_get_arch_string(int);
706extern const char *fio_get_os_string(int);
707
708#endif