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