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