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