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