projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Add iolog defines
[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 * roundrobin available files, or choose one at random, or do each one
378 * serially.
379 */
380enum {
381 FIO_FSERVICE_RANDOM = 1,
382 FIO_FSERVICE_RR = 2,
383 FIO_FSERVICE_SEQ = 3,
384};
385
386/*
387 * when should interactive ETA output be generated
388 */
389enum {
390 FIO_ETA_AUTO,
391 FIO_ETA_ALWAYS,
392 FIO_ETA_NEVER,
393};
394
395#define __td_verror(td, err, msg, func) \
396 do { \
397 if ((td)->error) \
398 break; \
399 int e = (err); \
400 (td)->error = e; \
401 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
402 } while (0)
403
404
405#define td_verror(td, err, func) \
406 __td_verror((td), (err), strerror((err)), (func))
407#define td_vmsg(td, err, msg, func) \
408 __td_verror((td), (err), (msg), (func))
409
410extern int exitall_on_terminate;
411extern int thread_number;
412extern int nr_process, nr_thread;
413extern int shm_id;
414extern int groupid;
415extern int terse_output;
416extern int temp_stall_ts;
417extern unsigned long long mlock_size;
418extern unsigned long page_mask, page_size;
419extern int read_only;
420extern int eta_print;
421extern unsigned long done_secs;
422extern char *job_section;
423extern int fio_gtod_offload;
424extern int fio_gtod_cpu;
425
426extern struct thread_data *threads;
427
428#define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
429#define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
430#define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
431#define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
432#define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
433
434static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
435{
436 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
437}
438
439#define BLOCKS_PER_MAP (8 * sizeof(int))
440#define TO_MAP_BLOCK(f, b) (b)
441#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
442#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
443
444#define MAX_JOBS (1024)
445
446static inline int should_fsync(struct thread_data *td)
447{
448 if (td->last_was_sync)
449 return 0;
450 if (td->o.odirect)
451 return 0;
452 if (td_write(td) || td_rw(td) || td->o.override_sync)
453 return 1;
454
455 return 0;
456}
457
458/*
459 * Disk utils as read in /sys/block/<dev>/stat
460 */
461struct disk_util_stat {
462 unsigned ios[2];
463 unsigned merges[2];
464 unsigned long long sectors[2];
465 unsigned ticks[2];
466 unsigned io_ticks;
467 unsigned time_in_queue;
468};
469
470/*
471 * Per-device disk util management
472 */
473struct disk_util {
474 struct flist_head list;
475 /* If this disk is a slave, hook it into the master's
476 * list using this head.
477 */
478 struct flist_head slavelist;
479
480 char *name;
481 char *sysfs_root;
482 char path[256];
483 int major, minor;
484
485 struct disk_util_stat dus;
486 struct disk_util_stat last_dus;
487
488 /* For software raids, this entry maintains pointers to the
489 * entries for the slave devices. The disk_util entries for
490 * the slaves devices should primarily be maintained through
491 * the disk_list list, i.e. for memory allocation and
492 * de-allocation, etc. Whereas this list should be used only
493 * for aggregating a software RAID's disk util figures.
494 */
495 struct flist_head slaves;
496
497 unsigned long msec;
498 struct timeval time;
499
500 struct fio_mutex *lock;
501 unsigned long users;
502};
503
504static inline void disk_util_inc(struct disk_util *du)
505{
506 if (du) {
507 fio_mutex_down(du->lock);
508 du->users++;
509 fio_mutex_up(du->lock);
510 }
511}
512
513static inline void disk_util_dec(struct disk_util *du)
514{
515 if (du) {
516 fio_mutex_down(du->lock);
517 du->users--;
518 fio_mutex_up(du->lock);
519 }
520}
521
522#define DISK_UTIL_MSEC (250)
523
524/*
525 * Time functions
526 */
527extern unsigned long long utime_since(struct timeval *, struct timeval *);
528extern unsigned long long utime_since_now(struct timeval *);
529extern unsigned long mtime_since(struct timeval *, struct timeval *);
530extern unsigned long mtime_since_now(struct timeval *);
531extern unsigned long time_since_now(struct timeval *);
532extern unsigned long mtime_since_genesis(void);
533extern void usec_spin(unsigned int);
534extern void usec_sleep(struct thread_data *, unsigned long);
535extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
536extern void fill_start_time(struct timeval *);
537extern void fio_gettime(struct timeval *, void *);
538extern void fio_gtod_init(void);
539extern void fio_gtod_update(void);
540extern void set_genesis_time(void);
541extern int ramp_time_over(struct thread_data *);
542extern int in_ramp_time(struct thread_data *);
543
544/*
545 * Init/option functions
546 */
547extern int __must_check parse_options(int, char **);
548extern int fio_options_parse(struct thread_data *, char **, int);
549extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
550extern void fio_fill_default_options(struct thread_data *);
551extern int fio_show_option_help(const char *);
552extern void fio_options_dup_and_init(struct option *);
553extern void options_mem_dupe(struct thread_data *);
554extern void options_mem_free(struct thread_data *);
555extern void td_fill_rand_seeds(struct thread_data *);
556#define FIO_GETOPT_JOB 0x89988998
557#define FIO_NR_OPTIONS 128
558
559/*
560 * ETA/status stuff
561 */
562extern void print_thread_status(void);
563extern void print_status_init(int);
564
565/*
566 * disk util stuff
567 */
568#ifdef FIO_HAVE_DISK_UTIL
569extern void show_disk_util(void);
570extern void init_disk_util(struct thread_data *);
571extern void update_io_ticks(void);
572#else
573#define show_disk_util()
574#define init_disk_util(td)
575#define update_io_ticks()
576#endif
577
578/*
579 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
580 * will never back again. It may cycle between running/verififying/fsyncing.
581 * Once the thread reaches TD_EXITED, it is just waiting for the core to
582 * reap it.
583 */
584enum {
585 TD_NOT_CREATED = 0,
586 TD_CREATED,
587 TD_INITIALIZED,
588 TD_RAMP,
589 TD_RUNNING,
590 TD_PRE_READING,
591 TD_VERIFYING,
592 TD_FSYNCING,
593 TD_EXITED,
594 TD_REAPED,
595};
596
597extern void td_set_runstate(struct thread_data *, int);
598
599/*
600 * Memory helpers
601 */
602extern int __must_check fio_pin_memory(void);
603extern void fio_unpin_memory(void);
604extern int __must_check allocate_io_mem(struct thread_data *);
605extern void free_io_mem(struct thread_data *);
606
607/*
608 * Reset stats after ramp time completes
609 */
610extern void reset_all_stats(struct thread_data *);
611
612/*
613 * blktrace support
614 */
615#ifdef FIO_HAVE_BLKTRACE
616extern int is_blktrace(const char *);
617extern int load_blktrace(struct thread_data *, const char *);
618#endif
619
620/*
621 * Mark unused variables passed to ops functions as unused, to silence gcc
622 */
623#define fio_unused __attribute((__unused__))
624#define fio_init __attribute__((constructor))
625#define fio_exit __attribute__((destructor))
626
627#define for_each_td(td, i) \
628 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
629#define for_each_file(td, f, i) \
630 if ((td)->files_index) \
631 for ((i) = 0, (f) = (td)->files[0]; \
632 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
633 (i)++)
634
635#define fio_assert(td, cond) do { \
636 if (!(cond)) { \
637 int *__foo = NULL; \
638 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
639 (td)->runstate = TD_EXITED; \
640 (td)->error = EFAULT; \
641 *__foo = 0; \
642 } \
643} while (0)
644
645static inline void fio_file_reset(struct fio_file *f)
646{
647 f->last_free_lookup = 0;
648 f->last_pos = f->file_offset;
649 if (f->file_map)
650 memset(f->file_map, 0, f->num_maps * sizeof(int));
651}
652
653static inline void clear_error(struct thread_data *td)
654{
655 td->error = 0;
656 td->verror[0] = '\0';
657}
658
659#ifdef FIO_INC_DEBUG
660static inline void dprint_io_u(struct io_u *io_u, const char *p)
661{
662 struct fio_file *f = io_u->file;
663
664 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
665 (unsigned long long) io_u->offset,
666 io_u->buflen, io_u->ddir);
667 if (fio_debug & (1 << FD_IO)) {
668 if (f)
669 log_info("/%s", f->file_name);
670
671 log_info("\n");
672 }
673}
674#else
675#define dprint_io_u(io_u, p)
676#endif
677
678static inline int fio_fill_issue_time(struct thread_data *td)
679{
680 if (td->o.read_iolog_file ||
681 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
682 return 1;
683
684 return 0;
685}
686
687#endif
fio - Flexible IO Tester