projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Revert "Improvements for num2str()"
[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 <inttypes.h>
15#include <assert.h>
16
17#include "compiler/compiler.h"
18#include "flist.h"
19#include "fifo.h"
20#include "rbtree.h"
21#include "arch/arch.h"
22#include "os/os.h"
23#include "mutex.h"
24#include "log.h"
25#include "debug.h"
26#include "file.h"
27#include "io_ddir.h"
28#include "ioengine.h"
29#include "iolog.h"
30#include "helpers.h"
31#include "options.h"
32#include "profile.h"
33#include "time.h"
34#include "lib/getopt.h"
35
36#ifdef FIO_HAVE_GUASI
37#include <guasi.h>
38#endif
39
40#ifdef FIO_HAVE_SOLARISAIO
41#include <sys/asynch.h>
42#endif
43
44struct group_run_stats {
45 unsigned long long max_run[2], min_run[2];
46 unsigned long long max_bw[2], min_bw[2];
47 unsigned long long io_kb[2];
48 unsigned long long agg[2];
49 unsigned int kb_base;
50};
51
52/*
53 * What type of allocation to use for io buffers
54 */
55enum fio_memtype {
56 MEM_MALLOC = 0, /* ordinary malloc */
57 MEM_SHM, /* use shared memory segments */
58 MEM_SHMHUGE, /* use shared memory segments with huge pages */
59 MEM_MMAP, /* use anonynomous mmap */
60 MEM_MMAPHUGE, /* memory mapped huge file */
61};
62
63/*
64 * offset generator types
65 */
66enum {
67 RW_SEQ_SEQ = 0,
68 RW_SEQ_IDENT,
69};
70
71/*
72 * How many depth levels to log
73 */
74#define FIO_IO_U_MAP_NR 8
75#define FIO_IO_U_LAT_U_NR 10
76#define FIO_IO_U_LAT_M_NR 12
77
78#define MAX_PATTERN_SIZE 512
79
80struct thread_stat {
81 char *name;
82 char *verror;
83 int error;
84 int groupid;
85 pid_t pid;
86 char *description;
87 int members;
88
89 struct io_log *slat_log;
90 struct io_log *clat_log;
91 struct io_log *lat_log;
92 struct io_log *bw_log;
93
94 /*
95 * bandwidth and latency stats
96 */
97 struct io_stat clat_stat[2]; /* completion latency */
98 struct io_stat slat_stat[2]; /* submission latency */
99 struct io_stat lat_stat[2]; /* total latency */
100 struct io_stat bw_stat[2]; /* bandwidth stats */
101
102 unsigned long long stat_io_bytes[2];
103 struct timeval stat_sample_time[2];
104
105 /*
106 * fio system usage accounting
107 */
108 struct rusage ru_start;
109 struct rusage ru_end;
110 unsigned long usr_time;
111 unsigned long sys_time;
112 unsigned long ctx;
113 unsigned long minf, majf;
114
115 /*
116 * IO depth and latency stats
117 */
118 unsigned int io_u_map[FIO_IO_U_MAP_NR];
119 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
120 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
121 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
122 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
123 unsigned long total_io_u[3];
124 unsigned long short_io_u[3];
125 unsigned long total_submit;
126 unsigned long total_complete;
127
128 unsigned long long io_bytes[2];
129 unsigned long runtime[2];
130 unsigned long total_run_time;
131
132 /*
133 * IO Error related stats
134 */
135 unsigned continue_on_error;
136 unsigned long total_err_count;
137 int first_error;
138
139 unsigned int kb_base;
140};
141
142struct bssplit {
143 unsigned int bs;
144 unsigned char perc;
145};
146
147struct thread_options {
148 int pad;
149 char *description;
150 char *name;
151 char *directory;
152 char *filename;
153 char *opendir;
154 char *ioengine;
155 enum td_ddir td_ddir;
156 unsigned int rw_seq;
157 unsigned int kb_base;
158 unsigned int ddir_seq_nr;
159 unsigned int iodepth;
160 unsigned int iodepth_low;
161 unsigned int iodepth_batch;
162 unsigned int iodepth_batch_complete;
163
164 unsigned long long size;
165 unsigned int fill_device;
166 unsigned long long file_size_low;
167 unsigned long long file_size_high;
168 unsigned long long start_offset;
169
170 unsigned int bs[2];
171 unsigned int ba[2];
172 unsigned int min_bs[2];
173 unsigned int max_bs[2];
174 struct bssplit *bssplit[2];
175 unsigned int bssplit_nr[2];
176
177 unsigned int nr_files;
178 unsigned int open_files;
179 enum file_lock_mode file_lock_mode;
180 unsigned int lockfile_batch;
181
182 unsigned int odirect;
183 unsigned int invalidate_cache;
184 unsigned int create_serialize;
185 unsigned int create_fsync;
186 unsigned int create_on_open;
187 unsigned int end_fsync;
188 unsigned int pre_read;
189 unsigned int sync_io;
190 unsigned int verify;
191 unsigned int do_verify;
192 unsigned int verifysort;
193 unsigned int verify_interval;
194 unsigned int verify_offset;
195 char verify_pattern[MAX_PATTERN_SIZE];
196 unsigned int verify_pattern_bytes;
197 unsigned int verify_fatal;
198 unsigned int verify_async;
199 unsigned long long verify_backlog;
200 unsigned int verify_batch;
201 unsigned int use_thread;
202 unsigned int unlink;
203 unsigned int do_disk_util;
204 unsigned int override_sync;
205 unsigned int rand_repeatable;
206 unsigned int write_lat_log;
207 unsigned int write_bw_log;
208 unsigned int norandommap;
209 unsigned int softrandommap;
210 unsigned int bs_unaligned;
211 unsigned int fsync_on_close;
212
213 unsigned int hugepage_size;
214 unsigned int rw_min_bs;
215 unsigned int thinktime;
216 unsigned int thinktime_spin;
217 unsigned int thinktime_blocks;
218 unsigned int fsync_blocks;
219 unsigned int fdatasync_blocks;
220 unsigned long start_delay;
221 unsigned long long timeout;
222 unsigned long long ramp_time;
223 unsigned int overwrite;
224 unsigned int bw_avg_time;
225 unsigned int loops;
226 unsigned long long zone_size;
227 unsigned long long zone_skip;
228 enum fio_memtype mem_type;
229 unsigned int mem_align;
230
231 unsigned int stonewall;
232 unsigned int new_group;
233 unsigned int numjobs;
234 os_cpu_mask_t cpumask;
235 unsigned int cpumask_set;
236 os_cpu_mask_t verify_cpumask;
237 unsigned int verify_cpumask_set;
238 unsigned int iolog;
239 unsigned int rwmixcycle;
240 unsigned int rwmix[2];
241 unsigned int nice;
242 unsigned int file_service_type;
243 unsigned int group_reporting;
244 unsigned int fadvise_hint;
245 unsigned int fallocate;
246 unsigned int zero_buffers;
247 unsigned int refill_buffers;
248 unsigned int time_based;
249 unsigned int disable_lat;
250 unsigned int disable_clat;
251 unsigned int disable_slat;
252 unsigned int disable_bw;
253 unsigned int gtod_reduce;
254 unsigned int gtod_cpu;
255 unsigned int gtod_offload;
256 enum fio_cs clocksource;
257 unsigned int no_stall;
258 unsigned int trim_percentage;
259 unsigned int trim_batch;
260 unsigned int trim_zero;
261 unsigned long long trim_backlog;
262
263 char *read_iolog_file;
264 char *write_iolog_file;
265 char *bw_log_file;
266 char *lat_log_file;
267 char *replay_redirect;
268
269 /*
270 * Pre-run and post-run shell
271 */
272 char *exec_prerun;
273 char *exec_postrun;
274
275 unsigned int rate[2];
276 unsigned int ratemin[2];
277 unsigned int ratecycle;
278 unsigned int rate_iops[2];
279 unsigned int rate_iops_min[2];
280
281 char *ioscheduler;
282
283 /*
284 * CPU "io" cycle burner
285 */
286 unsigned int cpuload;
287 unsigned int cpucycle;
288
289 /*
290 * I/O Error handling
291 */
292 unsigned int continue_on_error;
293
294 /*
295 * Benchmark profile type
296 */
297 char *profile;
298
299 /*
300 * blkio cgroup support
301 */
302 char *cgroup;
303 unsigned int cgroup_weight;
304 unsigned int cgroup_nodelete;
305
306 unsigned int uid;
307 unsigned int gid;
308
309 unsigned int sync_file_range;
310};
311
312#define FIO_VERROR_SIZE 128
313
314/*
315 * This describes a single thread/process executing a fio job.
316 */
317struct thread_data {
318 struct thread_options o;
319 char verror[FIO_VERROR_SIZE];
320 pthread_t thread;
321 int thread_number;
322 int groupid;
323 struct thread_stat ts;
324 struct fio_file **files;
325 unsigned int files_size;
326 unsigned int files_index;
327 unsigned int nr_open_files;
328 unsigned int nr_done_files;
329 unsigned int nr_normal_files;
330 union {
331 unsigned int next_file;
332 os_random_state_t next_file_state;
333 };
334 int error;
335 int done;
336 pid_t pid;
337 char *orig_buffer;
338 size_t orig_buffer_size;
339 volatile int terminate;
340 volatile int runstate;
341 unsigned int ioprio;
342 unsigned int ioprio_set;
343 unsigned int last_was_sync;
344 enum fio_ddir last_ddir;
345
346 char *mmapfile;
347 int mmapfd;
348
349 void *iolog_buf;
350 FILE *iolog_f;
351
352 char *sysfs_root;
353
354 unsigned long rand_seeds[7];
355
356 os_random_state_t bsrange_state;
357 os_random_state_t verify_state;
358 os_random_state_t trim_state;
359
360 unsigned int verify_batch;
361 unsigned int trim_batch;
362
363 int shm_id;
364
365 /*
366 * IO engine hooks, contains everything needed to submit an io_u
367 * to any of the available IO engines.
368 */
369 struct ioengine_ops *io_ops;
370
371 /*
372 * Current IO depth and list of free and busy io_u's.
373 */
374 unsigned int cur_depth;
375 unsigned int io_u_queued;
376 struct flist_head io_u_freelist;
377 struct flist_head io_u_busylist;
378 struct flist_head io_u_requeues;
379 pthread_mutex_t io_u_lock;
380 pthread_cond_t free_cond;
381
382 /*
383 * async verify offload
384 */
385 struct flist_head verify_list;
386 pthread_t *verify_threads;
387 unsigned int nr_verify_threads;
388 pthread_cond_t verify_cond;
389 int verify_thread_exit;
390
391 /*
392 * Rate state
393 */
394 unsigned long rate_nsec_cycle[2];
395 long rate_pending_usleep[2];
396 unsigned long rate_bytes[2];
397 unsigned long rate_blocks[2];
398 struct timeval lastrate[2];
399
400 unsigned long long total_io_size;
401 unsigned long long fill_device_size;
402
403 unsigned long io_issues[2];
404 unsigned long long io_blocks[2];
405 unsigned long long io_bytes[2];
406 unsigned long long io_skip_bytes;
407 unsigned long long this_io_bytes[2];
408 unsigned long long zone_bytes;
409 struct fio_mutex *mutex;
410
411 /*
412 * State for random io, a bitmap of blocks done vs not done
413 */
414 os_random_state_t random_state;
415
416 struct timeval start; /* start of this loop */
417 struct timeval epoch; /* time job was started */
418 struct timeval last_issue;
419 struct timeval tv_cache;
420 unsigned int tv_cache_nr;
421 unsigned int tv_cache_mask;
422 unsigned int ramp_time_over;
423
424 /*
425 * read/write mixed workload state
426 */
427 os_random_state_t rwmix_state;
428 unsigned long rwmix_issues;
429 enum fio_ddir rwmix_ddir;
430 unsigned int ddir_seq_nr;
431
432 /*
433 * IO history logs for verification. We use a tree for sorting,
434 * if we are overwriting. Otherwise just use a fifo.
435 */
436 struct rb_root io_hist_tree;
437 struct flist_head io_hist_list;
438 unsigned long io_hist_len;
439
440 /*
441 * For IO replaying
442 */
443 struct flist_head io_log_list;
444
445 /*
446 * For tracking/handling discards
447 */
448 struct flist_head trim_list;
449 unsigned long trim_entries;
450
451 /*
452 * for fileservice, how often to switch to a new file
453 */
454 unsigned int file_service_nr;
455 unsigned int file_service_left;
456 struct fio_file *file_service_file;
457
458 unsigned int sync_file_range_nr;
459
460 /*
461 * For generating file sizes
462 */
463 os_random_state_t file_size_state;
464
465 /*
466 * Error counts
467 */
468 unsigned int total_err_count;
469 int first_error;
470
471 /*
472 * Can be overloaded by profiles
473 */
474 struct prof_io_ops prof_io_ops;
475 void *prof_data;
476};
477
478/*
479 * when should interactive ETA output be generated
480 */
481enum {
482 FIO_ETA_AUTO,
483 FIO_ETA_ALWAYS,
484 FIO_ETA_NEVER,
485};
486
487#define __td_verror(td, err, msg, func) \
488 do { \
489 if ((td)->error) \
490 break; \
491 int e = (err); \
492 (td)->error = e; \
493 if (!(td)->first_error) \
494 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
495 } while (0)
496
497
498#define td_clear_error(td) \
499 (td)->error = 0;
500#define td_verror(td, err, func) \
501 __td_verror((td), (err), strerror((err)), (func))
502#define td_vmsg(td, err, msg, func) \
503 __td_verror((td), (err), (msg), (func))
504
505extern int exitall_on_terminate;
506extern int thread_number;
507extern int nr_process, nr_thread;
508extern int shm_id;
509extern int groupid;
510extern int terse_output;
511extern int temp_stall_ts;
512extern unsigned long long mlock_size;
513extern unsigned long page_mask, page_size;
514extern int read_only;
515extern int eta_print;
516extern unsigned long done_secs;
517extern char *job_section;
518extern int fio_gtod_offload;
519extern int fio_gtod_cpu;
520extern enum fio_cs fio_clock_source;
521
522extern struct thread_data *threads;
523
524static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
525{
526 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
527}
528
529#define BLOCKS_PER_MAP (8 * sizeof(int))
530#define TO_MAP_BLOCK(f, b) (b)
531#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
532#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
533
534#define MAX_JOBS (1024)
535
536#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
537
538static inline void update_error_count(struct thread_data *td, int err)
539{
540 td->total_err_count++;
541 if (td->total_err_count == 1)
542 td->first_error = err;
543}
544
545static inline int should_fsync(struct thread_data *td)
546{
547 if (td->last_was_sync)
548 return 0;
549 if (td->o.odirect)
550 return 0;
551 if (td_write(td) || td_rw(td) || td->o.override_sync)
552 return 1;
553
554 return 0;
555}
556
557/*
558 * Init/option functions
559 */
560extern int __must_check parse_options(int, char **);
561extern int fio_options_parse(struct thread_data *, char **, int);
562extern void fio_keywords_init(void);
563extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
564extern void fio_fill_default_options(struct thread_data *);
565extern int fio_show_option_help(const char *);
566extern void fio_options_dup_and_init(struct option *);
567extern void options_mem_dupe(struct thread_data *);
568extern void options_mem_free(struct thread_data *);
569extern void td_fill_rand_seeds(struct thread_data *);
570extern void add_job_opts(const char **);
571#define FIO_GETOPT_JOB 0x89988998
572#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
573
574/*
575 * ETA/status stuff
576 */
577extern void print_thread_status(void);
578extern void print_status_init(int);
579
580/*
581 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
582 * will never back again. It may cycle between running/verififying/fsyncing.
583 * Once the thread reaches TD_EXITED, it is just waiting for the core to
584 * reap it.
585 */
586enum {
587 TD_NOT_CREATED = 0,
588 TD_CREATED,
589 TD_INITIALIZED,
590 TD_RAMP,
591 TD_RUNNING,
592 TD_PRE_READING,
593 TD_VERIFYING,
594 TD_FSYNCING,
595 TD_EXITED,
596 TD_REAPED,
597};
598
599extern void td_set_runstate(struct thread_data *, int);
600
601/*
602 * Memory helpers
603 */
604extern int __must_check fio_pin_memory(void);
605extern void fio_unpin_memory(void);
606extern int __must_check allocate_io_mem(struct thread_data *);
607extern void free_io_mem(struct thread_data *);
608
609/*
610 * Reset stats after ramp time completes
611 */
612extern void reset_all_stats(struct thread_data *);
613
614/*
615 * blktrace support
616 */
617#ifdef FIO_HAVE_BLKTRACE
618extern int is_blktrace(const char *);
619extern int load_blktrace(struct thread_data *, const char *);
620#endif
621
622/*
623 * Mark unused variables passed to ops functions as unused, to silence gcc
624 */
625#define fio_unused __attribute((__unused__))
626#define fio_init __attribute__((constructor))
627#define fio_exit __attribute__((destructor))
628
629#define for_each_td(td, i) \
630 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
631#define for_each_file(td, f, i) \
632 if ((td)->files_index) \
633 for ((i) = 0, (f) = (td)->files[0]; \
634 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
635 (i)++)
636
637#define fio_assert(td, cond) do { \
638 if (!(cond)) { \
639 int *__foo = NULL; \
640 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
641 (td)->runstate = TD_EXITED; \
642 (td)->error = EFAULT; \
643 *__foo = 0; \
644 } \
645} while (0)
646
647static inline int fio_fill_issue_time(struct thread_data *td)
648{
649 if (td->o.read_iolog_file ||
650 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
651 return 1;
652
653 return 0;
654}
655
656/*
657 * Cheesy number->string conversion, complete with carry rounding error.
658 */
659static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
660{
661 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
662 unsigned int thousand;
663 char *buf;
664 int i;
665
666 if (pow2)
667 thousand = 1024;
668 else
669 thousand = 1000;
670
671 buf = malloc(128);
672
673 for (i = 0; base > 1; i++)
674 base /= thousand;
675
676 do {
677 int len, carry = 0;
678
679 len = sprintf(buf, "%'lu", num);
680 if (len <= maxlen) {
681 if (i >= 1) {
682 buf[len] = postfix[i];
683 buf[len + 1] = '\0';
684 }
685 return buf;
686 }
687
688 if ((num % thousand) >= (thousand / 2))
689 carry = 1;
690
691 num /= thousand;
692 num += carry;
693 i++;
694 } while (i <= 5);
695
696 return buf;
697}
698
699static inline int __should_check_rate(struct thread_data *td,
700 enum fio_ddir ddir)
701{
702 struct thread_options *o = &td->o;
703
704 /*
705 * If some rate setting was given, we need to check it
706 */
707 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
708 o->rate_iops_min[ddir])
709 return 1;
710
711 return 0;
712}
713
714static inline int should_check_rate(struct thread_data *td,
715 unsigned long *bytes_done)
716{
717 int ret = 0;
718
719 if (bytes_done[0])
720 ret |= __should_check_rate(td, 0);
721 if (bytes_done[1])
722 ret |= __should_check_rate(td, 1);
723
724 return ret;
725}
726
727static inline int is_power_of_2(unsigned int val)
728{
729 return (val != 0 && ((val & (val - 1)) == 0));
730}
731
732/*
733 * We currently only need to do locking if we have verifier threads
734 * accessing our internal structures too
735 */
736static inline void td_io_u_lock(struct thread_data *td)
737{
738 if (td->o.verify_async)
739 pthread_mutex_lock(&td->io_u_lock);
740}
741
742static inline void td_io_u_unlock(struct thread_data *td)
743{
744 if (td->o.verify_async)
745 pthread_mutex_unlock(&td->io_u_lock);
746}
747
748static inline void td_io_u_free_notify(struct thread_data *td)
749{
750 if (td->o.verify_async)
751 pthread_cond_signal(&td->free_cond);
752}
753
754#endif
fio - Flexible IO Tester