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