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