projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Move file service type defines to file.h
[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
116struct bssplit {
117 unsigned int bs;
118 unsigned char perc;
119};
120
121struct thread_options {
122 int pad;
123 char *description;
124 char *name;
125 char *directory;
126 char *filename;
127 char *opendir;
128 char *ioengine;
129 enum td_ddir td_ddir;
130 unsigned int ddir_nr;
131 unsigned int iodepth;
132 unsigned int iodepth_low;
133 unsigned int iodepth_batch;
134 unsigned int iodepth_batch_complete;
135
136 unsigned long long size;
137 unsigned int fill_device;
138 unsigned long long file_size_low;
139 unsigned long long file_size_high;
140 unsigned long long start_offset;
141
142 unsigned int bs[2];
143 unsigned int ba[2];
144 unsigned int min_bs[2];
145 unsigned int max_bs[2];
146 struct bssplit *bssplit[2];
147 unsigned int bssplit_nr[2];
148
149 unsigned int nr_files;
150 unsigned int open_files;
151 enum file_lock_mode file_lock_mode;
152 unsigned int lockfile_batch;
153
154 unsigned int odirect;
155 unsigned int invalidate_cache;
156 unsigned int create_serialize;
157 unsigned int create_fsync;
158 unsigned int create_on_open;
159 unsigned int end_fsync;
160 unsigned int pre_read;
161 unsigned int sync_io;
162 unsigned int verify;
163 unsigned int do_verify;
164 unsigned int verifysort;
165 unsigned int verify_interval;
166 unsigned int verify_offset;
167 unsigned int verify_pattern;
168 unsigned int verify_pattern_bytes;
169 unsigned int verify_fatal;
170 unsigned int use_thread;
171 unsigned int unlink;
172 unsigned int do_disk_util;
173 unsigned int override_sync;
174 unsigned int rand_repeatable;
175 unsigned int write_lat_log;
176 unsigned int write_bw_log;
177 unsigned int norandommap;
178 unsigned int softrandommap;
179 unsigned int bs_unaligned;
180 unsigned int fsync_on_close;
181
182 unsigned int hugepage_size;
183 unsigned int rw_min_bs;
184 unsigned int thinktime;
185 unsigned int thinktime_spin;
186 unsigned int thinktime_blocks;
187 unsigned int fsync_blocks;
188 unsigned int start_delay;
189 unsigned long long timeout;
190 unsigned long long ramp_time;
191 unsigned int overwrite;
192 unsigned int bw_avg_time;
193 unsigned int loops;
194 unsigned long long zone_size;
195 unsigned long long zone_skip;
196 enum fio_memtype mem_type;
197
198 unsigned int stonewall;
199 unsigned int new_group;
200 unsigned int numjobs;
201 os_cpu_mask_t cpumask;
202 unsigned int cpumask_set;
203 unsigned int iolog;
204 unsigned int rwmixcycle;
205 unsigned int rwmix[2];
206 unsigned int nice;
207 unsigned int file_service_type;
208 unsigned int group_reporting;
209 unsigned int fadvise_hint;
210 unsigned int zero_buffers;
211 unsigned int refill_buffers;
212 unsigned int time_based;
213 unsigned int disable_clat;
214 unsigned int disable_slat;
215 unsigned int disable_bw;
216 unsigned int gtod_reduce;
217 unsigned int gtod_cpu;
218 unsigned int gtod_offload;
219
220 char *read_iolog_file;
221 char *write_iolog_file;
222 char *bw_log_file;
223 char *lat_log_file;
224
225 /*
226 * Pre-run and post-run shell
227 */
228 char *exec_prerun;
229 char *exec_postrun;
230
231 unsigned int rate;
232 unsigned int ratemin;
233 unsigned int ratecycle;
234 unsigned int rate_iops;
235 unsigned int rate_iops_min;
236
237 char *ioscheduler;
238
239 /*
240 * CPU "io" cycle burner
241 */
242 unsigned int cpuload;
243 unsigned int cpucycle;
244};
245
246#define FIO_VERROR_SIZE 128
247
248/*
249 * This describes a single thread/process executing a fio job.
250 */
251struct thread_data {
252 struct thread_options o;
253 char verror[FIO_VERROR_SIZE];
254 pthread_t thread;
255 int thread_number;
256 int groupid;
257 struct thread_stat ts;
258 struct fio_file **files;
259 unsigned int files_size;
260 unsigned int files_index;
261 unsigned int nr_open_files;
262 unsigned int nr_done_files;
263 unsigned int nr_normal_files;
264 union {
265 unsigned int next_file;
266 os_random_state_t next_file_state;
267 };
268 int error;
269 int done;
270 pid_t pid;
271 char *orig_buffer;
272 size_t orig_buffer_size;
273 volatile int terminate;
274 volatile int runstate;
275 unsigned int ioprio;
276 unsigned int ioprio_set;
277 unsigned int last_was_sync;
278
279 char *mmapfile;
280 int mmapfd;
281
282 void *iolog_buf;
283 FILE *iolog_f;
284
285 char *sysfs_root;
286
287 unsigned long rand_seeds[6];
288
289 os_random_state_t bsrange_state;
290 os_random_state_t verify_state;
291
292 int shm_id;
293
294 /*
295 * IO engine hooks, contains everything needed to submit an io_u
296 * to any of the available IO engines.
297 */
298 struct ioengine_ops *io_ops;
299
300 /*
301 * Current IO depth and list of free and busy io_u's.
302 */
303 unsigned int cur_depth;
304 unsigned int io_u_queued;
305 struct flist_head io_u_freelist;
306 struct flist_head io_u_busylist;
307 struct flist_head io_u_requeues;
308
309 /*
310 * Rate state
311 */
312 unsigned long rate_usec_cycle;
313 long rate_pending_usleep;
314 unsigned long rate_bytes;
315 unsigned long rate_blocks;
316 struct timeval lastrate;
317
318 unsigned long long total_io_size;
319
320 unsigned long io_issues[2];
321 unsigned long long io_blocks[2];
322 unsigned long long io_bytes[2];
323 unsigned long long io_skip_bytes;
324 unsigned long long this_io_bytes[2];
325 unsigned long long zone_bytes;
326 struct fio_mutex *mutex;
327
328 /*
329 * State for random io, a bitmap of blocks done vs not done
330 */
331 os_random_state_t random_state;
332
333 struct timeval start; /* start of this loop */
334 struct timeval epoch; /* time job was started */
335 struct timeval rw_end[2];
336 struct timeval last_issue;
337 struct timeval tv_cache;
338 unsigned int tv_cache_nr;
339 unsigned int tv_cache_mask;
340 unsigned int rw_end_set[2];
341 unsigned int ramp_time_over;
342
343 /*
344 * read/write mixed workload state
345 */
346 os_random_state_t rwmix_state;
347 unsigned long rwmix_issues;
348 enum fio_ddir rwmix_ddir;
349 unsigned int ddir_nr;
350
351 /*
352 * IO history logs for verification. We use a tree for sorting,
353 * if we are overwriting. Otherwise just use a fifo.
354 */
355 struct rb_root io_hist_tree;
356 struct flist_head io_hist_list;
357
358 /*
359 * For IO replaying
360 */
361 struct flist_head io_log_list;
362
363 /*
364 * for fileservice, how often to switch to a new file
365 */
366 unsigned int file_service_nr;
367 unsigned int file_service_left;
368 struct fio_file *file_service_file;
369
370 /*
371 * For generating file sizes
372 */
373 os_random_state_t file_size_state;
374};
375
376/*
377 * when should interactive ETA output be generated
378 */
379enum {
380 FIO_ETA_AUTO,
381 FIO_ETA_ALWAYS,
382 FIO_ETA_NEVER,
383};
384
385#define __td_verror(td, err, msg, func) \
386 do { \
387 if ((td)->error) \
388 break; \
389 int e = (err); \
390 (td)->error = e; \
391 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
392 } while (0)
393
394
395#define td_verror(td, err, func) \
396 __td_verror((td), (err), strerror((err)), (func))
397#define td_vmsg(td, err, msg, func) \
398 __td_verror((td), (err), (msg), (func))
399
400extern int exitall_on_terminate;
401extern int thread_number;
402extern int nr_process, nr_thread;
403extern int shm_id;
404extern int groupid;
405extern int terse_output;
406extern int temp_stall_ts;
407extern unsigned long long mlock_size;
408extern unsigned long page_mask, page_size;
409extern int read_only;
410extern int eta_print;
411extern unsigned long done_secs;
412extern char *job_section;
413extern int fio_gtod_offload;
414extern int fio_gtod_cpu;
415
416extern struct thread_data *threads;
417
418#define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
419#define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
420#define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
421#define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
422#define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
423
424static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
425{
426 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
427}
428
429#define BLOCKS_PER_MAP (8 * sizeof(int))
430#define TO_MAP_BLOCK(f, b) (b)
431#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
432#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
433
434#define MAX_JOBS (1024)
435
436static inline int should_fsync(struct thread_data *td)
437{
438 if (td->last_was_sync)
439 return 0;
440 if (td->o.odirect)
441 return 0;
442 if (td_write(td) || td_rw(td) || td->o.override_sync)
443 return 1;
444
445 return 0;
446}
447
448/*
449 * Disk utils as read in /sys/block/<dev>/stat
450 */
451struct disk_util_stat {
452 unsigned ios[2];
453 unsigned merges[2];
454 unsigned long long sectors[2];
455 unsigned ticks[2];
456 unsigned io_ticks;
457 unsigned time_in_queue;
458};
459
460/*
461 * Per-device disk util management
462 */
463struct disk_util {
464 struct flist_head list;
465 /* If this disk is a slave, hook it into the master's
466 * list using this head.
467 */
468 struct flist_head slavelist;
469
470 char *name;
471 char *sysfs_root;
472 char path[256];
473 int major, minor;
474
475 struct disk_util_stat dus;
476 struct disk_util_stat last_dus;
477
478 /* For software raids, this entry maintains pointers to the
479 * entries for the slave devices. The disk_util entries for
480 * the slaves devices should primarily be maintained through
481 * the disk_list list, i.e. for memory allocation and
482 * de-allocation, etc. Whereas this list should be used only
483 * for aggregating a software RAID's disk util figures.
484 */
485 struct flist_head slaves;
486
487 unsigned long msec;
488 struct timeval time;
489
490 struct fio_mutex *lock;
491 unsigned long users;
492};
493
494static inline void disk_util_inc(struct disk_util *du)
495{
496 if (du) {
497 fio_mutex_down(du->lock);
498 du->users++;
499 fio_mutex_up(du->lock);
500 }
501}
502
503static inline void disk_util_dec(struct disk_util *du)
504{
505 if (du) {
506 fio_mutex_down(du->lock);
507 du->users--;
508 fio_mutex_up(du->lock);
509 }
510}
511
512#define DISK_UTIL_MSEC (250)
513
514/*
515 * Time functions
516 */
517extern unsigned long long utime_since(struct timeval *, struct timeval *);
518extern unsigned long long utime_since_now(struct timeval *);
519extern unsigned long mtime_since(struct timeval *, struct timeval *);
520extern unsigned long mtime_since_now(struct timeval *);
521extern unsigned long time_since_now(struct timeval *);
522extern unsigned long mtime_since_genesis(void);
523extern void usec_spin(unsigned int);
524extern void usec_sleep(struct thread_data *, unsigned long);
525extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
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 int fio_cmd_option_parse(struct thread_data *, const char *, char *);
540extern void fio_fill_default_options(struct thread_data *);
541extern int fio_show_option_help(const char *);
542extern void fio_options_dup_and_init(struct option *);
543extern void options_mem_dupe(struct thread_data *);
544extern void options_mem_free(struct thread_data *);
545extern void td_fill_rand_seeds(struct thread_data *);
546#define FIO_GETOPT_JOB 0x89988998
547#define FIO_NR_OPTIONS 128
548
549/*
550 * ETA/status stuff
551 */
552extern void print_thread_status(void);
553extern void print_status_init(int);
554
555/*
556 * disk util stuff
557 */
558#ifdef FIO_HAVE_DISK_UTIL
559extern void show_disk_util(void);
560extern void init_disk_util(struct thread_data *);
561extern void update_io_ticks(void);
562#else
563#define show_disk_util()
564#define init_disk_util(td)
565#define update_io_ticks()
566#endif
567
568/*
569 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
570 * will never back again. It may cycle between running/verififying/fsyncing.
571 * Once the thread reaches TD_EXITED, it is just waiting for the core to
572 * reap it.
573 */
574enum {
575 TD_NOT_CREATED = 0,
576 TD_CREATED,
577 TD_INITIALIZED,
578 TD_RAMP,
579 TD_RUNNING,
580 TD_PRE_READING,
581 TD_VERIFYING,
582 TD_FSYNCING,
583 TD_EXITED,
584 TD_REAPED,
585};
586
587extern void td_set_runstate(struct thread_data *, int);
588
589/*
590 * Memory helpers
591 */
592extern int __must_check fio_pin_memory(void);
593extern void fio_unpin_memory(void);
594extern int __must_check allocate_io_mem(struct thread_data *);
595extern void free_io_mem(struct thread_data *);
596
597/*
598 * Reset stats after ramp time completes
599 */
600extern void reset_all_stats(struct thread_data *);
601
602/*
603 * blktrace support
604 */
605#ifdef FIO_HAVE_BLKTRACE
606extern int is_blktrace(const char *);
607extern int load_blktrace(struct thread_data *, const char *);
608#endif
609
610/*
611 * Mark unused variables passed to ops functions as unused, to silence gcc
612 */
613#define fio_unused __attribute((__unused__))
614#define fio_init __attribute__((constructor))
615#define fio_exit __attribute__((destructor))
616
617#define for_each_td(td, i) \
618 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
619#define for_each_file(td, f, i) \
620 if ((td)->files_index) \
621 for ((i) = 0, (f) = (td)->files[0]; \
622 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
623 (i)++)
624
625#define fio_assert(td, cond) do { \
626 if (!(cond)) { \
627 int *__foo = NULL; \
628 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
629 (td)->runstate = TD_EXITED; \
630 (td)->error = EFAULT; \
631 *__foo = 0; \
632 } \
633} while (0)
634
635static inline void fio_file_reset(struct fio_file *f)
636{
637 f->last_free_lookup = 0;
638 f->last_pos = f->file_offset;
639 if (f->file_map)
640 memset(f->file_map, 0, f->num_maps * sizeof(int));
641}
642
643static inline void clear_error(struct thread_data *td)
644{
645 td->error = 0;
646 td->verror[0] = '\0';
647}
648
649#ifdef FIO_INC_DEBUG
650static inline void dprint_io_u(struct io_u *io_u, const char *p)
651{
652 struct fio_file *f = io_u->file;
653
654 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
655 (unsigned long long) io_u->offset,
656 io_u->buflen, io_u->ddir);
657 if (fio_debug & (1 << FD_IO)) {
658 if (f)
659 log_info("/%s", f->file_name);
660
661 log_info("\n");
662 }
663}
664#else
665#define dprint_io_u(io_u, p)
666#endif
667
668static inline int fio_fill_issue_time(struct thread_data *td)
669{
670 if (td->o.read_iolog_file ||
671 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
672 return 1;
673
674 return 0;
675}
676
677#endif
fio - Flexible IO Tester