Allow the adding of 'posval' for dynamic options like 'profile'
[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 "helpers.h"
32#include "options.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 fallocate;
231 unsigned int zero_buffers;
232 unsigned int refill_buffers;
233 unsigned int time_based;
234 unsigned int disable_clat;
235 unsigned int disable_slat;
236 unsigned int disable_bw;
237 unsigned int gtod_reduce;
238 unsigned int gtod_cpu;
239 unsigned int gtod_offload;
240
241 char *read_iolog_file;
242 char *write_iolog_file;
243 char *bw_log_file;
244 char *lat_log_file;
245
246 /*
247 * Pre-run and post-run shell
248 */
249 char *exec_prerun;
250 char *exec_postrun;
251
252 unsigned int rate[2];
253 unsigned int ratemin[2];
254 unsigned int ratecycle;
255 unsigned int rate_iops[2];
256 unsigned int rate_iops_min[2];
257
258 char *ioscheduler;
259
260 /*
261 * CPU "io" cycle burner
262 */
263 unsigned int cpuload;
264 unsigned int cpucycle;
265
266 /*
267 * I/O Error handling
268 */
269 unsigned int continue_on_error;
270
271 /*
272 * Benchmark profile type
273 */
274 char *profile;
275
276 /*
277 * blkio cgroup support
278 */
279 char *cgroup;
280 unsigned int cgroup_weight;
281
282 unsigned int uid;
283 unsigned int gid;
284};
285
286#define FIO_VERROR_SIZE 128
287
288/*
289 * This describes a single thread/process executing a fio job.
290 */
291struct thread_data {
292 struct thread_options o;
293 char verror[FIO_VERROR_SIZE];
294 pthread_t thread;
295 int thread_number;
296 int groupid;
297 struct thread_stat ts;
298 struct fio_file **files;
299 unsigned int files_size;
300 unsigned int files_index;
301 unsigned int nr_open_files;
302 unsigned int nr_done_files;
303 unsigned int nr_normal_files;
304 union {
305 unsigned int next_file;
306 os_random_state_t next_file_state;
307 };
308 int error;
309 int done;
310 pid_t pid;
311 char *orig_buffer;
312 size_t orig_buffer_size;
313 volatile int terminate;
314 volatile int runstate;
315 unsigned int ioprio;
316 unsigned int ioprio_set;
317 unsigned int last_was_sync;
318
319 char *mmapfile;
320 int mmapfd;
321
322 void *iolog_buf;
323 FILE *iolog_f;
324
325 char *sysfs_root;
326
327 unsigned long rand_seeds[6];
328
329 os_random_state_t bsrange_state;
330 os_random_state_t verify_state;
331
332 int shm_id;
333
334 /*
335 * IO engine hooks, contains everything needed to submit an io_u
336 * to any of the available IO engines.
337 */
338 struct ioengine_ops *io_ops;
339
340 /*
341 * Current IO depth and list of free and busy io_u's.
342 */
343 unsigned int cur_depth;
344 unsigned int io_u_queued;
345 struct flist_head io_u_freelist;
346 struct flist_head io_u_busylist;
347 struct flist_head io_u_requeues;
348 pthread_mutex_t io_u_lock;
349 pthread_cond_t free_cond;
350
351 /*
352 * async verify offload
353 */
354 struct flist_head verify_list;
355 pthread_t *verify_threads;
356 unsigned int nr_verify_threads;
357 pthread_cond_t verify_cond;
358 int verify_thread_exit;
359
360 /*
361 * Rate state
362 */
363 unsigned long rate_nsec_cycle[2];
364 long rate_pending_usleep[2];
365 unsigned long rate_bytes[2];
366 unsigned long rate_blocks[2];
367 struct timeval lastrate[2];
368
369 unsigned long long total_io_size;
370
371 unsigned long io_issues[2];
372 unsigned long long io_blocks[2];
373 unsigned long long io_bytes[2];
374 unsigned long long io_skip_bytes;
375 unsigned long long this_io_bytes[2];
376 unsigned long long zone_bytes;
377 struct fio_mutex *mutex;
378
379 /*
380 * State for random io, a bitmap of blocks done vs not done
381 */
382 os_random_state_t random_state;
383
384 struct timeval start; /* start of this loop */
385 struct timeval epoch; /* time job was started */
386 struct timeval last_issue;
387 struct timeval tv_cache;
388 unsigned int tv_cache_nr;
389 unsigned int tv_cache_mask;
390 unsigned int ramp_time_over;
391
392 /*
393 * read/write mixed workload state
394 */
395 os_random_state_t rwmix_state;
396 unsigned long rwmix_issues;
397 enum fio_ddir rwmix_ddir;
398 unsigned int ddir_nr;
399
400 /*
401 * IO history logs for verification. We use a tree for sorting,
402 * if we are overwriting. Otherwise just use a fifo.
403 */
404 struct rb_root io_hist_tree;
405 struct flist_head io_hist_list;
406
407 /*
408 * For IO replaying
409 */
410 struct flist_head io_log_list;
411
412 /*
413 * for fileservice, how often to switch to a new file
414 */
415 unsigned int file_service_nr;
416 unsigned int file_service_left;
417 struct fio_file *file_service_file;
418
419 /*
420 * For generating file sizes
421 */
422 os_random_state_t file_size_state;
423
424 /*
425 * Error counts
426 */
427 unsigned int total_err_count;
428 int first_error;
429};
430
431/*
432 * when should interactive ETA output be generated
433 */
434enum {
435 FIO_ETA_AUTO,
436 FIO_ETA_ALWAYS,
437 FIO_ETA_NEVER,
438};
439
440#define __td_verror(td, err, msg, func) \
441 do { \
442 if ((td)->error) \
443 break; \
444 int e = (err); \
445 (td)->error = e; \
446 if (!(td)->first_error) \
447 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
448 } while (0)
449
450
451#define td_clear_error(td) \
452 (td)->error = 0;
453#define td_verror(td, err, func) \
454 __td_verror((td), (err), strerror((err)), (func))
455#define td_vmsg(td, err, msg, func) \
456 __td_verror((td), (err), (msg), (func))
457
458extern int exitall_on_terminate;
459extern int thread_number;
460extern int nr_process, nr_thread;
461extern int shm_id;
462extern int groupid;
463extern int terse_output;
464extern int temp_stall_ts;
465extern unsigned long long mlock_size;
466extern unsigned long page_mask, page_size;
467extern int read_only;
468extern int eta_print;
469extern unsigned long done_secs;
470extern char *job_section;
471extern int fio_gtod_offload;
472extern int fio_gtod_cpu;
473
474extern struct thread_data *threads;
475
476static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
477{
478 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
479}
480
481#define BLOCKS_PER_MAP (8 * sizeof(int))
482#define TO_MAP_BLOCK(f, b) (b)
483#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
484#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
485
486#define MAX_JOBS (1024)
487
488#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
489
490static inline void update_error_count(struct thread_data *td, int err)
491{
492 td->total_err_count++;
493 if (td->total_err_count == 1)
494 td->first_error = err;
495}
496
497static inline int should_fsync(struct thread_data *td)
498{
499 if (td->last_was_sync)
500 return 0;
501 if (td->o.odirect)
502 return 0;
503 if (td_write(td) || td_rw(td) || td->o.override_sync)
504 return 1;
505
506 return 0;
507}
508
509/*
510 * Time functions
511 */
512extern unsigned long long utime_since(struct timeval *, struct timeval *);
513extern unsigned long long utime_since_now(struct timeval *);
514extern unsigned long mtime_since(struct timeval *, struct timeval *);
515extern unsigned long mtime_since_now(struct timeval *);
516extern unsigned long time_since_now(struct timeval *);
517extern unsigned long mtime_since_genesis(void);
518extern void usec_spin(unsigned int);
519extern void usec_sleep(struct thread_data *, unsigned long);
520extern void fill_start_time(struct timeval *);
521extern void fio_gettime(struct timeval *, void *);
522extern void fio_gtod_init(void);
523extern void fio_gtod_update(void);
524extern void set_genesis_time(void);
525extern int ramp_time_over(struct thread_data *);
526extern int in_ramp_time(struct thread_data *);
527
528/*
529 * Init/option functions
530 */
531extern int __must_check parse_options(int, char **);
532extern int fio_options_parse(struct thread_data *, char **, int);
533extern void fio_keywords_init(void);
534extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
535extern void fio_fill_default_options(struct thread_data *);
536extern int fio_show_option_help(const char *);
537extern void fio_options_dup_and_init(struct option *);
538extern void options_mem_dupe(struct thread_data *);
539extern void options_mem_free(struct thread_data *);
540extern void td_fill_rand_seeds(struct thread_data *);
541extern void add_job_opts(const char **);
542#define FIO_GETOPT_JOB 0x89988998
543#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
544
545/*
546 * ETA/status stuff
547 */
548extern void print_thread_status(void);
549extern void print_status_init(int);
550
551/*
552 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
553 * will never back again. It may cycle between running/verififying/fsyncing.
554 * Once the thread reaches TD_EXITED, it is just waiting for the core to
555 * reap it.
556 */
557enum {
558 TD_NOT_CREATED = 0,
559 TD_CREATED,
560 TD_INITIALIZED,
561 TD_RAMP,
562 TD_RUNNING,
563 TD_PRE_READING,
564 TD_VERIFYING,
565 TD_FSYNCING,
566 TD_EXITED,
567 TD_REAPED,
568};
569
570extern void td_set_runstate(struct thread_data *, int);
571
572/*
573 * Memory helpers
574 */
575extern int __must_check fio_pin_memory(void);
576extern void fio_unpin_memory(void);
577extern int __must_check allocate_io_mem(struct thread_data *);
578extern void free_io_mem(struct thread_data *);
579
580/*
581 * Reset stats after ramp time completes
582 */
583extern void reset_all_stats(struct thread_data *);
584
585/*
586 * blktrace support
587 */
588#ifdef FIO_HAVE_BLKTRACE
589extern int is_blktrace(const char *);
590extern int load_blktrace(struct thread_data *, const char *);
591#endif
592
593/*
594 * Mark unused variables passed to ops functions as unused, to silence gcc
595 */
596#define fio_unused __attribute((__unused__))
597#define fio_init __attribute__((constructor))
598#define fio_exit __attribute__((destructor))
599
600#define for_each_td(td, i) \
601 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
602#define for_each_file(td, f, i) \
603 if ((td)->files_index) \
604 for ((i) = 0, (f) = (td)->files[0]; \
605 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
606 (i)++)
607
608#define fio_assert(td, cond) do { \
609 if (!(cond)) { \
610 int *__foo = NULL; \
611 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
612 (td)->runstate = TD_EXITED; \
613 (td)->error = EFAULT; \
614 *__foo = 0; \
615 } \
616} while (0)
617
618static inline int fio_fill_issue_time(struct thread_data *td)
619{
620 if (td->o.read_iolog_file ||
621 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
622 return 1;
623
624 return 0;
625}
626
627/*
628 * Cheesy number->string conversion, complete with carry rounding error.
629 */
630static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
631{
632 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
633 unsigned int thousand;
634 char *buf;
635 int i;
636
637 if (pow2)
638 thousand = 1024;
639 else
640 thousand = 1000;
641
642 buf = malloc(128);
643
644 for (i = 0; base > 1; i++)
645 base /= thousand;
646
647 do {
648 int len, carry = 0;
649
650 len = sprintf(buf, "%'lu", num);
651 if (len <= maxlen) {
652 if (i >= 1) {
653 buf[len] = postfix[i];
654 buf[len + 1] = '\0';
655 }
656 return buf;
657 }
658
659 if ((num % thousand) >= (thousand / 2))
660 carry = 1;
661
662 num /= thousand;
663 num += carry;
664 i++;
665 } while (i <= 5);
666
667 return buf;
668}
669
670static inline int __should_check_rate(struct thread_data *td,
671 enum fio_ddir ddir)
672{
673 struct thread_options *o = &td->o;
674
675 /*
676 * If some rate setting was given, we need to check it
677 */
678 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
679 o->rate_iops_min[ddir])
680 return 1;
681
682 return 0;
683}
684
685static inline int should_check_rate(struct thread_data *td,
686 unsigned long *bytes_done)
687{
688 int ret = 0;
689
690 if (bytes_done[0])
691 ret |= __should_check_rate(td, 0);
692 if (bytes_done[1])
693 ret |= __should_check_rate(td, 1);
694
695 return ret;
696}
697
698static inline int is_power_of_2(unsigned int val)
699{
700 return (val != 0 && ((val & (val - 1)) == 0));
701}
702
703/*
704 * We currently only need to do locking if we have verifier threads
705 * accessing our internal structures too
706 */
707static inline void td_io_u_lock(struct thread_data *td)
708{
709 if (td->o.verify_async)
710 pthread_mutex_lock(&td->io_u_lock);
711}
712
713static inline void td_io_u_unlock(struct thread_data *td)
714{
715 if (td->o.verify_async)
716 pthread_mutex_unlock(&td->io_u_lock);
717}
718
719static inline void td_io_u_free_notify(struct thread_data *td)
720{
721 if (td->o.verify_async)
722 pthread_cond_signal(&td->free_cond);
723}
724
725#endif