projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge branch 'master' of ssh://brick.kernel.dk/data/git/fio
[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 "profiles.h"
32#include "helpers.h"
33
34#ifdef FIO_HAVE_GUASI
35#include <guasi.h>
36#endif
37
38#ifdef FIO_HAVE_SOLARISAIO
39#include <sys/asynch.h>
40#endif
41
42struct group_run_stats {
43 unsigned long long max_run[2], min_run[2];
44 unsigned long long max_bw[2], min_bw[2];
45 unsigned long long io_kb[2];
46 unsigned long long agg[2];
47 unsigned int kb_base;
48};
49
50/*
51 * What type of allocation to use for io buffers
52 */
53enum fio_memtype {
54 MEM_MALLOC = 0, /* ordinary malloc */
55 MEM_SHM, /* use shared memory segments */
56 MEM_SHMHUGE, /* use shared memory segments with huge pages */
57 MEM_MMAP, /* use anonynomous mmap */
58 MEM_MMAPHUGE, /* memory mapped huge file */
59};
60
61/*
62 * How many depth levels to log
63 */
64#define FIO_IO_U_MAP_NR 8
65#define FIO_IO_U_LAT_U_NR 10
66#define FIO_IO_U_LAT_M_NR 12
67
68#define MAX_PATTERN_SIZE 512
69
70struct thread_stat {
71 char *name;
72 char *verror;
73 int error;
74 int groupid;
75 pid_t pid;
76 char *description;
77 int members;
78
79 struct io_log *slat_log;
80 struct io_log *clat_log;
81 struct io_log *bw_log;
82
83 /*
84 * bandwidth and latency stats
85 */
86 struct io_stat clat_stat[2]; /* completion latency */
87 struct io_stat slat_stat[2]; /* submission latency */
88 struct io_stat bw_stat[2]; /* bandwidth stats */
89
90 unsigned long long stat_io_bytes[2];
91 struct timeval stat_sample_time[2];
92
93 /*
94 * fio system usage accounting
95 */
96 struct rusage ru_start;
97 struct rusage ru_end;
98 unsigned long usr_time;
99 unsigned long sys_time;
100 unsigned long ctx;
101 unsigned long minf, majf;
102
103 /*
104 * IO depth and latency stats
105 */
106 unsigned int io_u_map[FIO_IO_U_MAP_NR];
107 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
108 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
109 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
110 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
111 unsigned long total_io_u[2];
112 unsigned long short_io_u[2];
113 unsigned long total_submit;
114 unsigned long total_complete;
115
116 unsigned long long io_bytes[2];
117 unsigned long runtime[2];
118 unsigned long total_run_time;
119
120 /*
121 * IO Error related stats
122 */
123 unsigned continue_on_error;
124 unsigned long total_err_count;
125 int first_error;
126
127 unsigned int kb_base;
128};
129
130struct bssplit {
131 unsigned int bs;
132 unsigned char perc;
133};
134
135struct thread_options {
136 int pad;
137 char *description;
138 char *name;
139 char *directory;
140 char *filename;
141 char *opendir;
142 char *ioengine;
143 enum td_ddir td_ddir;
144 unsigned int kb_base;
145 unsigned int ddir_nr;
146 unsigned int iodepth;
147 unsigned int iodepth_low;
148 unsigned int iodepth_batch;
149 unsigned int iodepth_batch_complete;
150
151 unsigned long long size;
152 unsigned int fill_device;
153 unsigned long long file_size_low;
154 unsigned long long file_size_high;
155 unsigned long long start_offset;
156
157 unsigned int bs[2];
158 unsigned int ba[2];
159 unsigned int min_bs[2];
160 unsigned int max_bs[2];
161 struct bssplit *bssplit[2];
162 unsigned int bssplit_nr[2];
163
164 unsigned int nr_files;
165 unsigned int open_files;
166 enum file_lock_mode file_lock_mode;
167 unsigned int lockfile_batch;
168
169 unsigned int odirect;
170 unsigned int invalidate_cache;
171 unsigned int create_serialize;
172 unsigned int create_fsync;
173 unsigned int create_on_open;
174 unsigned int end_fsync;
175 unsigned int pre_read;
176 unsigned int sync_io;
177 unsigned int verify;
178 unsigned int do_verify;
179 unsigned int verifysort;
180 unsigned int verify_interval;
181 unsigned int verify_offset;
182 char verify_pattern[MAX_PATTERN_SIZE];
183 unsigned int verify_pattern_bytes;
184 unsigned int verify_fatal;
185 unsigned int verify_async;
186 unsigned int use_thread;
187 unsigned int unlink;
188 unsigned int do_disk_util;
189 unsigned int override_sync;
190 unsigned int rand_repeatable;
191 unsigned int write_lat_log;
192 unsigned int write_bw_log;
193 unsigned int norandommap;
194 unsigned int softrandommap;
195 unsigned int bs_unaligned;
196 unsigned int fsync_on_close;
197
198 unsigned int hugepage_size;
199 unsigned int rw_min_bs;
200 unsigned int thinktime;
201 unsigned int thinktime_spin;
202 unsigned int thinktime_blocks;
203 unsigned int fsync_blocks;
204 unsigned int fdatasync_blocks;
205 unsigned int start_delay;
206 unsigned long long timeout;
207 unsigned long long ramp_time;
208 unsigned int overwrite;
209 unsigned int bw_avg_time;
210 unsigned int loops;
211 unsigned long long zone_size;
212 unsigned long long zone_skip;
213 enum fio_memtype mem_type;
214 unsigned int mem_align;
215
216 unsigned int stonewall;
217 unsigned int new_group;
218 unsigned int numjobs;
219 os_cpu_mask_t cpumask;
220 unsigned int cpumask_set;
221 os_cpu_mask_t verify_cpumask;
222 unsigned int verify_cpumask_set;
223 unsigned int iolog;
224 unsigned int rwmixcycle;
225 unsigned int rwmix[2];
226 unsigned int nice;
227 unsigned int file_service_type;
228 unsigned int group_reporting;
229 unsigned int fadvise_hint;
230 unsigned int zero_buffers;
231 unsigned int refill_buffers;
232 unsigned int time_based;
233 unsigned int disable_clat;
234 unsigned int disable_slat;
235 unsigned int disable_bw;
236 unsigned int gtod_reduce;
237 unsigned int gtod_cpu;
238 unsigned int gtod_offload;
239
240 char *read_iolog_file;
241 char *write_iolog_file;
242 char *bw_log_file;
243 char *lat_log_file;
244
245 /*
246 * Pre-run and post-run shell
247 */
248 char *exec_prerun;
249 char *exec_postrun;
250
251 unsigned int rate[2];
252 unsigned int ratemin[2];
253 unsigned int ratecycle;
254 unsigned int rate_iops[2];
255 unsigned int rate_iops_min[2];
256
257 char *ioscheduler;
258
259 /*
260 * CPU "io" cycle burner
261 */
262 unsigned int cpuload;
263 unsigned int cpucycle;
264
265 /*
266 * I/O Error handling
267 */
268 unsigned int continue_on_error;
269
270 /*
271 * Benchmark profile type
272 */
273 unsigned int profile;
274
275 /*
276 * blkio cgroup support
277 */
278 char *cgroup;
279 unsigned int cgroup_weight;
280
281 unsigned int uid;
282 unsigned int gid;
283};
284
285#define FIO_VERROR_SIZE 128
286
287/*
288 * This describes a single thread/process executing a fio job.
289 */
290struct thread_data {
291 struct thread_options o;
292 char verror[FIO_VERROR_SIZE];
293 pthread_t thread;
294 int thread_number;
295 int groupid;
296 struct thread_stat ts;
297 struct fio_file **files;
298 unsigned int files_size;
299 unsigned int files_index;
300 unsigned int nr_open_files;
301 unsigned int nr_done_files;
302 unsigned int nr_normal_files;
303 union {
304 unsigned int next_file;
305 os_random_state_t next_file_state;
306 };
307 int error;
308 int done;
309 pid_t pid;
310 char *orig_buffer;
311 size_t orig_buffer_size;
312 volatile int terminate;
313 volatile int runstate;
314 unsigned int ioprio;
315 unsigned int ioprio_set;
316 unsigned int last_was_sync;
317
318 char *mmapfile;
319 int mmapfd;
320
321 void *iolog_buf;
322 FILE *iolog_f;
323
324 char *sysfs_root;
325
326 unsigned long rand_seeds[6];
327
328 os_random_state_t bsrange_state;
329 os_random_state_t verify_state;
330
331 int shm_id;
332
333 /*
334 * IO engine hooks, contains everything needed to submit an io_u
335 * to any of the available IO engines.
336 */
337 struct ioengine_ops *io_ops;
338
339 /*
340 * Current IO depth and list of free and busy io_u's.
341 */
342 unsigned int cur_depth;
343 unsigned int io_u_queued;
344 struct flist_head io_u_freelist;
345 struct flist_head io_u_busylist;
346 struct flist_head io_u_requeues;
347 pthread_mutex_t io_u_lock;
348 pthread_cond_t free_cond;
349
350 /*
351 * async verify offload
352 */
353 struct flist_head verify_list;
354 pthread_t *verify_threads;
355 unsigned int nr_verify_threads;
356 pthread_cond_t verify_cond;
357 int verify_thread_exit;
358
359 /*
360 * Rate state
361 */
362 unsigned long rate_nsec_cycle[2];
363 long rate_pending_usleep[2];
364 unsigned long rate_bytes[2];
365 unsigned long rate_blocks[2];
366 struct timeval lastrate[2];
367
368 unsigned long long total_io_size;
369
370 unsigned long io_issues[2];
371 unsigned long long io_blocks[2];
372 unsigned long long io_bytes[2];
373 unsigned long long io_skip_bytes;
374 unsigned long long this_io_bytes[2];
375 unsigned long long zone_bytes;
376 struct fio_mutex *mutex;
377
378 /*
379 * State for random io, a bitmap of blocks done vs not done
380 */
381 os_random_state_t random_state;
382
383 struct timeval start; /* start of this loop */
384 struct timeval epoch; /* time job was started */
385 struct timeval last_issue;
386 struct timeval tv_cache;
387 unsigned int tv_cache_nr;
388 unsigned int tv_cache_mask;
389 unsigned int ramp_time_over;
390
391 /*
392 * read/write mixed workload state
393 */
394 os_random_state_t rwmix_state;
395 unsigned long rwmix_issues;
396 enum fio_ddir rwmix_ddir;
397 unsigned int ddir_nr;
398
399 /*
400 * IO history logs for verification. We use a tree for sorting,
401 * if we are overwriting. Otherwise just use a fifo.
402 */
403 struct rb_root io_hist_tree;
404 struct flist_head io_hist_list;
405
406 /*
407 * For IO replaying
408 */
409 struct flist_head io_log_list;
410
411 /*
412 * for fileservice, how often to switch to a new file
413 */
414 unsigned int file_service_nr;
415 unsigned int file_service_left;
416 struct fio_file *file_service_file;
417
418 /*
419 * For generating file sizes
420 */
421 os_random_state_t file_size_state;
422
423 /*
424 * Error counts
425 */
426 unsigned int total_err_count;
427 int first_error;
428};
429
430/*
431 * when should interactive ETA output be generated
432 */
433enum {
434 FIO_ETA_AUTO,
435 FIO_ETA_ALWAYS,
436 FIO_ETA_NEVER,
437};
438
439#define __td_verror(td, err, msg, func) \
440 do { \
441 if ((td)->error) \
442 break; \
443 int e = (err); \
444 (td)->error = e; \
445 if (!(td)->first_error) \
446 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
447 } while (0)
448
449
450#define td_clear_error(td) \
451 (td)->error = 0;
452#define td_verror(td, err, func) \
453 __td_verror((td), (err), strerror((err)), (func))
454#define td_vmsg(td, err, msg, func) \
455 __td_verror((td), (err), (msg), (func))
456
457extern int exitall_on_terminate;
458extern int thread_number;
459extern int nr_process, nr_thread;
460extern int shm_id;
461extern int groupid;
462extern int terse_output;
463extern int temp_stall_ts;
464extern unsigned long long mlock_size;
465extern unsigned long page_mask, page_size;
466extern int read_only;
467extern int eta_print;
468extern unsigned long done_secs;
469extern char *job_section;
470extern int fio_gtod_offload;
471extern int fio_gtod_cpu;
472
473extern struct thread_data *threads;
474
475static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
476{
477 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
478}
479
480#define BLOCKS_PER_MAP (8 * sizeof(int))
481#define TO_MAP_BLOCK(f, b) (b)
482#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
483#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
484
485#define MAX_JOBS (1024)
486
487#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
488
489static inline void update_error_count(struct thread_data *td, int err)
490{
491 td->total_err_count++;
492 if (td->total_err_count == 1)
493 td->first_error = err;
494}
495
496static inline int should_fsync(struct thread_data *td)
497{
498 if (td->last_was_sync)
499 return 0;
500 if (td->o.odirect)
501 return 0;
502 if (td_write(td) || td_rw(td) || td->o.override_sync)
503 return 1;
504
505 return 0;
506}
507
508/*
509 * Time functions
510 */
511extern unsigned long long utime_since(struct timeval *, struct timeval *);
512extern unsigned long long utime_since_now(struct timeval *);
513extern unsigned long mtime_since(struct timeval *, struct timeval *);
514extern unsigned long mtime_since_now(struct timeval *);
515extern unsigned long time_since_now(struct timeval *);
516extern unsigned long mtime_since_genesis(void);
517extern void usec_spin(unsigned int);
518extern void usec_sleep(struct thread_data *, unsigned long);
519extern void fill_start_time(struct timeval *);
520extern void fio_gettime(struct timeval *, void *);
521extern void fio_gtod_init(void);
522extern void fio_gtod_update(void);
523extern void set_genesis_time(void);
524extern int ramp_time_over(struct thread_data *);
525extern int in_ramp_time(struct thread_data *);
526
527/*
528 * Init/option functions
529 */
530extern int __must_check parse_options(int, char **);
531extern int fio_options_parse(struct thread_data *, char **, int);
532extern void fio_keywords_init(void);
533extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
534extern void fio_fill_default_options(struct thread_data *);
535extern int fio_show_option_help(const char *);
536extern void fio_options_dup_and_init(struct option *);
537extern void options_mem_dupe(struct thread_data *);
538extern void options_mem_free(struct thread_data *);
539extern void td_fill_rand_seeds(struct thread_data *);
540#define FIO_GETOPT_JOB 0x89988998
541#define FIO_NR_OPTIONS 128
542
543/*
544 * ETA/status stuff
545 */
546extern void print_thread_status(void);
547extern void print_status_init(int);
548
549/*
550 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
551 * will never back again. It may cycle between running/verififying/fsyncing.
552 * Once the thread reaches TD_EXITED, it is just waiting for the core to
553 * reap it.
554 */
555enum {
556 TD_NOT_CREATED = 0,
557 TD_CREATED,
558 TD_INITIALIZED,
559 TD_RAMP,
560 TD_RUNNING,
561 TD_PRE_READING,
562 TD_VERIFYING,
563 TD_FSYNCING,
564 TD_EXITED,
565 TD_REAPED,
566};
567
568extern void td_set_runstate(struct thread_data *, int);
569
570/*
571 * Memory helpers
572 */
573extern int __must_check fio_pin_memory(void);
574extern void fio_unpin_memory(void);
575extern int __must_check allocate_io_mem(struct thread_data *);
576extern void free_io_mem(struct thread_data *);
577
578/*
579 * Reset stats after ramp time completes
580 */
581extern void reset_all_stats(struct thread_data *);
582
583/*
584 * blktrace support
585 */
586#ifdef FIO_HAVE_BLKTRACE
587extern int is_blktrace(const char *);
588extern int load_blktrace(struct thread_data *, const char *);
589#endif
590
591/*
592 * Mark unused variables passed to ops functions as unused, to silence gcc
593 */
594#define fio_unused __attribute((__unused__))
595#define fio_init __attribute__((constructor))
596#define fio_exit __attribute__((destructor))
597
598#define for_each_td(td, i) \
599 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
600#define for_each_file(td, f, i) \
601 if ((td)->files_index) \
602 for ((i) = 0, (f) = (td)->files[0]; \
603 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
604 (i)++)
605
606#define fio_assert(td, cond) do { \
607 if (!(cond)) { \
608 int *__foo = NULL; \
609 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
610 (td)->runstate = TD_EXITED; \
611 (td)->error = EFAULT; \
612 *__foo = 0; \
613 } \
614} while (0)
615
616static inline int fio_fill_issue_time(struct thread_data *td)
617{
618 if (td->o.read_iolog_file ||
619 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
620 return 1;
621
622 return 0;
623}
624
625/*
626 * Cheesy number->string conversion, complete with carry rounding error.
627 */
628static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
629{
630 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
631 unsigned int thousand;
632 char *buf;
633 int i;
634
635 if (pow2)
636 thousand = 1024;
637 else
638 thousand = 1000;
639
640 buf = malloc(128);
641
642 for (i = 0; base > 1; i++)
643 base /= thousand;
644
645 do {
646 int len, carry = 0;
647
648 len = sprintf(buf, "%'lu", num);
649 if (len <= maxlen) {
650 if (i >= 1) {
651 buf[len] = postfix[i];
652 buf[len + 1] = '\0';
653 }
654 return buf;
655 }
656
657 if ((num % thousand) >= (thousand / 2))
658 carry = 1;
659
660 num /= thousand;
661 num += carry;
662 i++;
663 } while (i <= 5);
664
665 return buf;
666}
667
668static inline int __should_check_rate(struct thread_data *td,
669 enum fio_ddir ddir)
670{
671 struct thread_options *o = &td->o;
672
673 /*
674 * If some rate setting was given, we need to check it
675 */
676 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
677 o->rate_iops_min[ddir])
678 return 1;
679
680 return 0;
681}
682
683static inline int should_check_rate(struct thread_data *td,
684 unsigned long *bytes_done)
685{
686 int ret = 0;
687
688 if (bytes_done[0])
689 ret |= __should_check_rate(td, 0);
690 if (bytes_done[1])
691 ret |= __should_check_rate(td, 1);
692
693 return ret;
694}
695
696static inline int is_power_of_2(unsigned int val)
697{
698 return (val != 0 && ((val & (val - 1)) == 0));
699}
700
701/*
702 * We currently only need to do locking if we have verifier threads
703 * accessing our internal structures too
704 */
705static inline void td_io_u_lock(struct thread_data *td)
706{
707 if (td->o.verify_async)
708 pthread_mutex_lock(&td->io_u_lock);
709}
710
711static inline void td_io_u_unlock(struct thread_data *td)
712{
713 if (td->o.verify_async)
714 pthread_mutex_unlock(&td->io_u_lock);
715}
716
717static inline void td_io_u_free_notify(struct thread_data *td)
718{
719 if (td->o.verify_async)
720 pthread_cond_signal(&td->free_cond);
721}
722
723#endif
fio - Flexible IO Tester