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