Mark buffer_compress_chunk as child of buffer_compress_percentage
[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
17struct thread_data;
18
19#include "compiler/compiler.h"
20#include "flist.h"
21#include "fifo.h"
22#include "rbtree.h"
23#include "arch/arch.h"
24#include "os/os.h"
25#include "mutex.h"
26#include "log.h"
27#include "debug.h"
28#include "file.h"
29#include "io_ddir.h"
30#include "ioengine.h"
31#include "iolog.h"
32#include "helpers.h"
33#include "options.h"
34#include "profile.h"
35#include "time.h"
36#include "gettime.h"
37#include "lib/getopt.h"
38#include "lib/rand.h"
39#include "server.h"
40#include "stat.h"
41#include "flow.h"
42
43#ifdef FIO_HAVE_GUASI
44#include <guasi.h>
45#endif
46
47#ifdef FIO_HAVE_SOLARISAIO
48#include <sys/asynch.h>
49#endif
50
51/*
52 * What type of allocation to use for io buffers
53 */
54enum fio_memtype {
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
60};
61
62/*
63 * offset generator types
64 */
65enum {
66 RW_SEQ_SEQ = 0,
67 RW_SEQ_IDENT,
68};
69
70/*
71 * What type of errors to continue on when continue_on_error is used
72 */
73enum error_type {
74 ERROR_TYPE_NONE = 0,
75 ERROR_TYPE_READ = 1 << 0,
76 ERROR_TYPE_WRITE = 1 << 1,
77 ERROR_TYPE_VERIFY = 1 << 2,
78 ERROR_TYPE_ANY = 0xffff,
79};
80
81struct bssplit {
82 unsigned int bs;
83 unsigned char perc;
84};
85
86struct thread_options {
87 int pad;
88 char *description;
89 char *name;
90 char *directory;
91 char *filename;
92 char *opendir;
93 char *ioengine;
94 enum td_ddir td_ddir;
95 unsigned int rw_seq;
96 unsigned int kb_base;
97 unsigned int ddir_seq_nr;
98 long ddir_seq_add;
99 unsigned int iodepth;
100 unsigned int iodepth_low;
101 unsigned int iodepth_batch;
102 unsigned int iodepth_batch_complete;
103
104 unsigned long long size;
105 unsigned int size_percent;
106 unsigned int fill_device;
107 unsigned long long file_size_low;
108 unsigned long long file_size_high;
109 unsigned long long start_offset;
110
111 unsigned int bs[2];
112 unsigned int ba[2];
113 unsigned int min_bs[2];
114 unsigned int max_bs[2];
115 struct bssplit *bssplit[2];
116 unsigned int bssplit_nr[2];
117
118 unsigned int nr_files;
119 unsigned int open_files;
120 enum file_lock_mode file_lock_mode;
121 unsigned int lockfile_batch;
122
123 unsigned int odirect;
124 unsigned int invalidate_cache;
125 unsigned int create_serialize;
126 unsigned int create_fsync;
127 unsigned int create_on_open;
128 unsigned int end_fsync;
129 unsigned int pre_read;
130 unsigned int sync_io;
131 unsigned int verify;
132 unsigned int do_verify;
133 unsigned int verifysort;
134 unsigned int verify_interval;
135 unsigned int verify_offset;
136 char verify_pattern[MAX_PATTERN_SIZE];
137 unsigned int verify_pattern_bytes;
138 unsigned int verify_fatal;
139 unsigned int verify_dump;
140 unsigned int verify_async;
141 unsigned long long verify_backlog;
142 unsigned int verify_batch;
143 unsigned int use_thread;
144 unsigned int unlink;
145 unsigned int do_disk_util;
146 unsigned int override_sync;
147 unsigned int rand_repeatable;
148 unsigned int use_os_rand;
149 unsigned int write_lat_log;
150 unsigned int write_bw_log;
151 unsigned int write_iops_log;
152 unsigned int log_avg_msec;
153 unsigned int norandommap;
154 unsigned int softrandommap;
155 unsigned int bs_unaligned;
156 unsigned int fsync_on_close;
157
158 unsigned int hugepage_size;
159 unsigned int rw_min_bs;
160 unsigned int thinktime;
161 unsigned int thinktime_spin;
162 unsigned int thinktime_blocks;
163 unsigned int fsync_blocks;
164 unsigned int fdatasync_blocks;
165 unsigned int barrier_blocks;
166 unsigned long long start_delay;
167 unsigned long long timeout;
168 unsigned long long ramp_time;
169 unsigned int overwrite;
170 unsigned int bw_avg_time;
171 unsigned int iops_avg_time;
172 unsigned int loops;
173 unsigned long long zone_range;
174 unsigned long long zone_size;
175 unsigned long long zone_skip;
176 enum fio_memtype mem_type;
177 unsigned int mem_align;
178
179 unsigned int stonewall;
180 unsigned int new_group;
181 unsigned int numjobs;
182 os_cpu_mask_t cpumask;
183 unsigned int cpumask_set;
184 os_cpu_mask_t verify_cpumask;
185 unsigned int verify_cpumask_set;
186 unsigned int iolog;
187 unsigned int rwmixcycle;
188 unsigned int rwmix[2];
189 unsigned int nice;
190 unsigned int file_service_type;
191 unsigned int group_reporting;
192 unsigned int fadvise_hint;
193 enum fio_fallocate_mode fallocate_mode;
194 unsigned int zero_buffers;
195 unsigned int refill_buffers;
196 unsigned int scramble_buffers;
197 unsigned int compress_percentage;
198 unsigned int compress_chunk;
199 unsigned int time_based;
200 unsigned int disable_lat;
201 unsigned int disable_clat;
202 unsigned int disable_slat;
203 unsigned int disable_bw;
204 unsigned int gtod_reduce;
205 unsigned int gtod_cpu;
206 unsigned int gtod_offload;
207 enum fio_cs clocksource;
208 unsigned int no_stall;
209 unsigned int trim_percentage;
210 unsigned int trim_batch;
211 unsigned int trim_zero;
212 unsigned long long trim_backlog;
213 unsigned int clat_percentiles;
214 unsigned int overwrite_plist;
215 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
216
217 char *read_iolog_file;
218 char *write_iolog_file;
219 char *bw_log_file;
220 char *lat_log_file;
221 char *iops_log_file;
222 char *replay_redirect;
223
224 /*
225 * Pre-run and post-run shell
226 */
227 char *exec_prerun;
228 char *exec_postrun;
229
230 unsigned int rate[2];
231 unsigned int ratemin[2];
232 unsigned int ratecycle;
233 unsigned int rate_iops[2];
234 unsigned int rate_iops_min[2];
235
236 char *ioscheduler;
237
238 /*
239 * CPU "io" cycle burner
240 */
241 unsigned int cpuload;
242 unsigned int cpucycle;
243
244 /*
245 * I/O Error handling
246 */
247 enum error_type continue_on_error;
248
249 /*
250 * Benchmark profile type
251 */
252 char *profile;
253
254 /*
255 * blkio cgroup support
256 */
257 char *cgroup;
258 unsigned int cgroup_weight;
259 unsigned int cgroup_nodelete;
260
261 unsigned int uid;
262 unsigned int gid;
263
264 int flow_id;
265 int flow;
266 int flow_watermark;
267 unsigned int flow_sleep;
268
269 unsigned int sync_file_range;
270};
271
272/*
273 * This describes a single thread/process executing a fio job.
274 */
275struct thread_data {
276 struct thread_options o;
277 void *eo;
278 char verror[FIO_VERROR_SIZE];
279 pthread_t thread;
280 int thread_number;
281 int groupid;
282 struct thread_stat ts;
283
284 struct io_log *slat_log;
285 struct io_log *clat_log;
286 struct io_log *lat_log;
287 struct io_log *bw_log;
288 struct io_log *iops_log;
289
290 uint64_t stat_io_bytes[2];
291 struct timeval bw_sample_time;
292
293 uint64_t stat_io_blocks[2];
294 struct timeval iops_sample_time;
295
296 struct rusage ru_start;
297 struct rusage ru_end;
298
299 struct fio_file **files;
300 unsigned int files_size;
301 unsigned int files_index;
302 unsigned int nr_open_files;
303 unsigned int nr_done_files;
304 unsigned int nr_normal_files;
305 union {
306 unsigned int next_file;
307 os_random_state_t next_file_state;
308 struct frand_state __next_file_state;
309 };
310 int error;
311 int done;
312 pid_t pid;
313 char *orig_buffer;
314 size_t orig_buffer_size;
315 volatile int terminate;
316 volatile int runstate;
317 unsigned int ioprio;
318 unsigned int ioprio_set;
319 unsigned int last_was_sync;
320 enum fio_ddir last_ddir;
321
322 char *mmapfile;
323 int mmapfd;
324
325 void *iolog_buf;
326 FILE *iolog_f;
327
328 char *sysfs_root;
329
330 unsigned long rand_seeds[8];
331
332 union {
333 os_random_state_t bsrange_state;
334 struct frand_state __bsrange_state;
335 };
336 union {
337 os_random_state_t verify_state;
338 struct frand_state __verify_state;
339 };
340 union {
341 os_random_state_t trim_state;
342 struct frand_state __trim_state;
343 };
344
345 struct frand_state buf_state;
346
347 unsigned int verify_batch;
348 unsigned int trim_batch;
349
350 int shm_id;
351
352 /*
353 * IO engine hooks, contains everything needed to submit an io_u
354 * to any of the available IO engines.
355 */
356 struct ioengine_ops *io_ops;
357
358 /*
359 * Queue depth of io_u's that fio MIGHT do
360 */
361 unsigned int cur_depth;
362
363 /*
364 * io_u's about to be committed
365 */
366 unsigned int io_u_queued;
367
368 /*
369 * io_u's submitted but not completed yet
370 */
371 unsigned int io_u_in_flight;
372
373 /*
374 * List of free and busy io_u's
375 */
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 long rate_bps[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 this_io_blocks[2];
406 unsigned long long io_bytes[2];
407 unsigned long long io_skip_bytes;
408 unsigned long long this_io_bytes[2];
409 unsigned long long zone_bytes;
410 struct fio_mutex *mutex;
411
412 /*
413 * State for random io, a bitmap of blocks done vs not done
414 */
415 union {
416 os_random_state_t random_state;
417 struct frand_state __random_state;
418 };
419
420 struct timeval start; /* start of this loop */
421 struct timeval epoch; /* time job was started */
422 struct timeval last_issue;
423 struct timeval tv_cache;
424 unsigned int tv_cache_nr;
425 unsigned int tv_cache_mask;
426 unsigned int ramp_time_over;
427
428 /*
429 * read/write mixed workload state
430 */
431 union {
432 os_random_state_t rwmix_state;
433 struct frand_state __rwmix_state;
434 };
435 unsigned long rwmix_issues;
436 enum fio_ddir rwmix_ddir;
437 unsigned int ddir_seq_nr;
438
439 /*
440 * IO history logs for verification. We use a tree for sorting,
441 * if we are overwriting. Otherwise just use a fifo.
442 */
443 struct rb_root io_hist_tree;
444 struct flist_head io_hist_list;
445 unsigned long io_hist_len;
446
447 /*
448 * For IO replaying
449 */
450 struct flist_head io_log_list;
451
452 /*
453 * For tracking/handling discards
454 */
455 struct flist_head trim_list;
456 unsigned long trim_entries;
457
458 /*
459 * for fileservice, how often to switch to a new file
460 */
461 unsigned int file_service_nr;
462 unsigned int file_service_left;
463 struct fio_file *file_service_file;
464
465 unsigned int sync_file_range_nr;
466
467 /*
468 * For generating file sizes
469 */
470 union {
471 os_random_state_t file_size_state;
472 struct frand_state __file_size_state;
473 };
474
475 /*
476 * Error counts
477 */
478 unsigned int total_err_count;
479 int first_error;
480
481 struct fio_flow *flow;
482
483 /*
484 * Can be overloaded by profiles
485 */
486 struct prof_io_ops prof_io_ops;
487 void *prof_data;
488};
489
490/*
491 * when should interactive ETA output be generated
492 */
493enum {
494 FIO_ETA_AUTO,
495 FIO_ETA_ALWAYS,
496 FIO_ETA_NEVER,
497};
498
499#define __td_verror(td, err, msg, func) \
500 do { \
501 if ((td)->error) \
502 break; \
503 int e = (err); \
504 (td)->error = e; \
505 if (!(td)->first_error) \
506 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
507 } while (0)
508
509
510#define td_clear_error(td) \
511 (td)->error = 0;
512#define td_verror(td, err, func) \
513 __td_verror((td), (err), strerror((err)), (func))
514#define td_vmsg(td, err, msg, func) \
515 __td_verror((td), (err), (msg), (func))
516
517#define __fio_stringify_1(x) #x
518#define __fio_stringify(x) __fio_stringify_1(x)
519
520extern int exitall_on_terminate;
521extern unsigned int thread_number;
522extern unsigned int nr_process, nr_thread;
523extern int shm_id;
524extern int groupid;
525extern int terse_output;
526extern int temp_stall_ts;
527extern unsigned long long mlock_size;
528extern unsigned long page_mask, page_size;
529extern int read_only;
530extern int eta_print;
531extern unsigned long done_secs;
532extern char *job_section;
533extern int fio_gtod_offload;
534extern int fio_gtod_cpu;
535extern enum fio_cs fio_clock_source;
536extern int warnings_fatal;
537extern int terse_version;
538extern int is_backend;
539extern int nr_clients;
540extern int log_syslog;
541extern const char fio_version_string[];
542extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
543
544extern struct thread_data *threads;
545
546static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
547{
548 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
549}
550
551#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
552#define TO_MAP_BLOCK(f, b) (b)
553#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
554#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
555
556#define REAL_MAX_JOBS 2048
557
558#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
559
560static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
561{
562 if (err == EILSEQ)
563 return ERROR_TYPE_VERIFY;
564 if (ddir == DDIR_READ)
565 return ERROR_TYPE_READ;
566 return ERROR_TYPE_WRITE;
567}
568
569static inline void update_error_count(struct thread_data *td, int err)
570{
571 td->total_err_count++;
572 if (td->total_err_count == 1)
573 td->first_error = err;
574}
575
576static inline int should_fsync(struct thread_data *td)
577{
578 if (td->last_was_sync)
579 return 0;
580 if (td->o.odirect)
581 return 0;
582 if (td_write(td) || td_rw(td) || td->o.override_sync)
583 return 1;
584
585 return 0;
586}
587
588/*
589 * Init/option functions
590 */
591extern int __must_check parse_options(int, char **);
592extern int parse_jobs_ini(char *, int, int);
593extern int parse_cmd_line(int, char **);
594extern int fio_backend(void);
595extern void reset_fio_state(void);
596extern void clear_io_state(struct thread_data *);
597extern int fio_options_parse(struct thread_data *, char **, int);
598extern void fio_keywords_init(void);
599extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
600extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
601extern void fio_fill_default_options(struct thread_data *);
602extern int fio_show_option_help(const char *);
603extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
604extern void fio_options_dup_and_init(struct option *);
605extern void fio_options_mem_dupe(struct thread_data *);
606extern void options_mem_dupe(void *data, struct fio_option *options);
607extern void td_fill_rand_seeds(struct thread_data *);
608extern void add_job_opts(const char **);
609extern char *num2str(unsigned long, int, int, int);
610extern int ioengine_load(struct thread_data *);
611
612#define FIO_GETOPT_JOB 0x89000000
613#define FIO_GETOPT_IOENGINE 0x98000000
614#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
615
616/*
617 * ETA/status stuff
618 */
619extern void print_thread_status(void);
620extern void print_status_init(int);
621
622/*
623 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
624 * will never back again. It may cycle between running/verififying/fsyncing.
625 * Once the thread reaches TD_EXITED, it is just waiting for the core to
626 * reap it.
627 */
628enum {
629 TD_NOT_CREATED = 0,
630 TD_CREATED,
631 TD_INITIALIZED,
632 TD_RAMP,
633 TD_RUNNING,
634 TD_PRE_READING,
635 TD_VERIFYING,
636 TD_FSYNCING,
637 TD_EXITED,
638 TD_REAPED,
639};
640
641extern void td_set_runstate(struct thread_data *, int);
642#define TERMINATE_ALL (-1)
643extern void fio_terminate_threads(int);
644
645/*
646 * Memory helpers
647 */
648extern int __must_check fio_pin_memory(void);
649extern void fio_unpin_memory(void);
650extern int __must_check allocate_io_mem(struct thread_data *);
651extern void free_io_mem(struct thread_data *);
652
653/*
654 * Reset stats after ramp time completes
655 */
656extern void reset_all_stats(struct thread_data *);
657
658/*
659 * blktrace support
660 */
661#ifdef FIO_HAVE_BLKTRACE
662extern int is_blktrace(const char *);
663extern int load_blktrace(struct thread_data *, const char *);
664#endif
665
666/*
667 * Mark unused variables passed to ops functions as unused, to silence gcc
668 */
669#define fio_unused __attribute((__unused__))
670#define fio_init __attribute__((constructor))
671#define fio_exit __attribute__((destructor))
672
673#define for_each_td(td, i) \
674 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
675#define for_each_file(td, f, i) \
676 if ((td)->files_index) \
677 for ((i) = 0, (f) = (td)->files[0]; \
678 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
679 (i)++)
680
681#define fio_assert(td, cond) do { \
682 if (!(cond)) { \
683 int *__foo = NULL; \
684 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
685 td_set_runstate((td), TD_EXITED); \
686 (td)->error = EFAULT; \
687 *__foo = 0; \
688 } \
689} while (0)
690
691static inline int fio_fill_issue_time(struct thread_data *td)
692{
693 if (td->o.read_iolog_file ||
694 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
695 return 1;
696
697 return 0;
698}
699
700static inline int __should_check_rate(struct thread_data *td,
701 enum fio_ddir ddir)
702{
703 struct thread_options *o = &td->o;
704
705 /*
706 * If some rate setting was given, we need to check it
707 */
708 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
709 o->rate_iops_min[ddir])
710 return 1;
711
712 return 0;
713}
714
715static inline int should_check_rate(struct thread_data *td,
716 unsigned long *bytes_done)
717{
718 int ret = 0;
719
720 if (bytes_done[0])
721 ret |= __should_check_rate(td, 0);
722 if (bytes_done[1])
723 ret |= __should_check_rate(td, 1);
724
725 return ret;
726}
727
728static inline int is_power_of_2(unsigned int val)
729{
730 return (val != 0 && ((val & (val - 1)) == 0));
731}
732
733/*
734 * We currently only need to do locking if we have verifier threads
735 * accessing our internal structures too
736 */
737static inline void td_io_u_lock(struct thread_data *td)
738{
739 if (td->o.verify_async)
740 pthread_mutex_lock(&td->io_u_lock);
741}
742
743static inline void td_io_u_unlock(struct thread_data *td)
744{
745 if (td->o.verify_async)
746 pthread_mutex_unlock(&td->io_u_lock);
747}
748
749static inline void td_io_u_free_notify(struct thread_data *td)
750{
751 if (td->o.verify_async)
752 pthread_cond_signal(&td->free_cond);
753}
754
755extern const char *fio_get_arch_string(int);
756extern const char *fio_get_os_string(int);
757
758#endif