projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Update documentation for random_distribution=gauss
[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 "thread_options.h"
19#include "flist.h"
20#include "fifo.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 "fio_time.h"
34#include "gettime.h"
35#include "oslib/getopt.h"
36#include "lib/rand.h"
37#include "lib/rbtree.h"
38#include "client.h"
39#include "server.h"
40#include "stat.h"
41#include "flow.h"
42#include "io_u_queue.h"
43#include "workqueue.h"
44
45#ifdef CONFIG_SOLARISAIO
46#include <sys/asynch.h>
47#endif
48
49#ifdef CONFIG_LIBNUMA
50#include <linux/mempolicy.h>
51#include <numa.h>
52
53/*
54 * "local" is pseudo-policy
55 */
56#define MPOL_LOCAL MPOL_MAX
57#endif
58
59/*
60 * offset generator types
61 */
62enum {
63 RW_SEQ_SEQ = 0,
64 RW_SEQ_IDENT,
65};
66
67enum {
68 TD_F_VER_BACKLOG = 1U << 0,
69 TD_F_TRIM_BACKLOG = 1U << 1,
70 TD_F_READ_IOLOG = 1U << 2,
71 TD_F_REFILL_BUFFERS = 1U << 3,
72 TD_F_SCRAMBLE_BUFFERS = 1U << 4,
73 TD_F_VER_NONE = 1U << 5,
74 TD_F_PROFILE_OPS = 1U << 6,
75 TD_F_COMPRESS = 1U << 7,
76 TD_F_NOIO = 1U << 8,
77 TD_F_COMPRESS_LOG = 1U << 9,
78 TD_F_VSTATE_SAVED = 1U << 10,
79 TD_F_NEED_LOCK = 1U << 11,
80 TD_F_CHILD = 1U << 12,
81 TD_F_NO_PROGRESS = 1U << 13,
82};
83
84enum {
85 FIO_RAND_BS_OFF = 0,
86 FIO_RAND_VER_OFF,
87 FIO_RAND_MIX_OFF,
88 FIO_RAND_FILE_OFF,
89 FIO_RAND_BLOCK_OFF,
90 FIO_RAND_FILE_SIZE_OFF,
91 FIO_RAND_TRIM_OFF,
92 FIO_RAND_BUF_OFF,
93 FIO_RAND_SEQ_RAND_READ_OFF,
94 FIO_RAND_SEQ_RAND_WRITE_OFF,
95 FIO_RAND_SEQ_RAND_TRIM_OFF,
96 FIO_RAND_START_DELAY,
97 FIO_DEDUPE_OFF,
98 FIO_RAND_POISSON_OFF,
99 FIO_RAND_NR_OFFS,
100};
101
102enum {
103 IO_MODE_INLINE = 0,
104 IO_MODE_OFFLOAD = 1,
105
106 RATE_PROCESS_LINEAR = 0,
107 RATE_PROCESS_POISSON = 1,
108};
109
110/*
111 * Per-thread/process specific data. Only used for the network client
112 * for now.
113 */
114struct sk_out;
115void sk_out_assign(struct sk_out *);
116void sk_out_drop(void);
117
118/*
119 * This describes a single thread/process executing a fio job.
120 */
121struct thread_data {
122 struct thread_options o;
123 struct flist_head opt_list;
124 unsigned long flags;
125 void *eo;
126 char verror[FIO_VERROR_SIZE];
127 pthread_t thread;
128 unsigned int thread_number;
129 unsigned int subjob_number;
130 unsigned int groupid;
131 struct thread_stat ts;
132
133 int client_type;
134
135 struct io_log *slat_log;
136 struct io_log *clat_log;
137 struct io_log *lat_log;
138 struct io_log *bw_log;
139 struct io_log *iops_log;
140
141 struct workqueue log_compress_wq;
142
143 struct thread_data *parent;
144
145 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
146 struct timeval bw_sample_time;
147
148 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
149 struct timeval iops_sample_time;
150
151 /*
152 * Tracks the last iodepth number of completed writes, if data
153 * verification is enabled
154 */
155 uint64_t *last_write_comp;
156 unsigned int last_write_idx;
157
158 volatile int update_rusage;
159 struct fio_mutex *rusage_sem;
160 struct rusage ru_start;
161 struct rusage ru_end;
162
163 struct fio_file **files;
164 unsigned char *file_locks;
165 unsigned int files_size;
166 unsigned int files_index;
167 unsigned int nr_open_files;
168 unsigned int nr_done_files;
169 unsigned int nr_normal_files;
170 union {
171 unsigned int next_file;
172 struct frand_state next_file_state;
173 };
174 int error;
175 int sig;
176 int done;
177 int stop_io;
178 pid_t pid;
179 char *orig_buffer;
180 size_t orig_buffer_size;
181 volatile int terminate;
182 volatile int runstate;
183 unsigned int last_was_sync;
184 enum fio_ddir last_ddir;
185
186 int mmapfd;
187
188 void *iolog_buf;
189 FILE *iolog_f;
190
191 char *sysfs_root;
192
193 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
194
195 struct frand_state bsrange_state;
196 struct frand_state verify_state;
197 struct frand_state trim_state;
198 struct frand_state delay_state;
199
200 struct frand_state buf_state;
201 struct frand_state buf_state_prev;
202 struct frand_state dedupe_state;
203
204 unsigned int verify_batch;
205 unsigned int trim_batch;
206
207 struct thread_io_list *vstate;
208
209 int shm_id;
210
211 /*
212 * IO engine hooks, contains everything needed to submit an io_u
213 * to any of the available IO engines.
214 */
215 struct ioengine_ops *io_ops;
216
217 /*
218 * Queue depth of io_u's that fio MIGHT do
219 */
220 unsigned int cur_depth;
221
222 /*
223 * io_u's about to be committed
224 */
225 unsigned int io_u_queued;
226
227 /*
228 * io_u's submitted but not completed yet
229 */
230 unsigned int io_u_in_flight;
231
232 /*
233 * List of free and busy io_u's
234 */
235 struct io_u_ring io_u_requeues;
236 struct io_u_queue io_u_freelist;
237 struct io_u_queue io_u_all;
238 pthread_mutex_t io_u_lock;
239 pthread_cond_t free_cond;
240
241 /*
242 * async verify offload
243 */
244 struct flist_head verify_list;
245 pthread_t *verify_threads;
246 unsigned int nr_verify_threads;
247 pthread_cond_t verify_cond;
248 int verify_thread_exit;
249
250 /*
251 * Rate state
252 */
253 uint64_t rate_bps[DDIR_RWDIR_CNT];
254 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
255 unsigned long rate_bytes[DDIR_RWDIR_CNT];
256 unsigned long rate_blocks[DDIR_RWDIR_CNT];
257 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
258 struct timeval lastrate[DDIR_RWDIR_CNT];
259 int64_t last_usec;
260 struct frand_state poisson_state;
261
262 /*
263 * Enforced rate submission/completion workqueue
264 */
265 struct workqueue io_wq;
266
267 uint64_t total_io_size;
268 uint64_t fill_device_size;
269
270 /*
271 * Issue side
272 */
273 uint64_t io_issues[DDIR_RWDIR_CNT];
274 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
275 uint64_t loops;
276
277 /*
278 * Completions
279 */
280 uint64_t io_blocks[DDIR_RWDIR_CNT];
281 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
282 uint64_t io_bytes[DDIR_RWDIR_CNT];
283 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
284 uint64_t io_skip_bytes;
285 uint64_t zone_bytes;
286 struct fio_mutex *mutex;
287 uint64_t bytes_done[DDIR_RWDIR_CNT];
288
289 /*
290 * State for random io, a bitmap of blocks done vs not done
291 */
292 struct frand_state random_state;
293
294 struct timeval start; /* start of this loop */
295 struct timeval epoch; /* time job was started */
296 struct timeval last_issue;
297 long time_offset;
298 struct timeval tv_cache;
299 struct timeval terminate_time;
300 unsigned int tv_cache_nr;
301 unsigned int tv_cache_mask;
302 unsigned int ramp_time_over;
303
304 /*
305 * Time since last latency_window was started
306 */
307 struct timeval latency_ts;
308 unsigned int latency_qd;
309 unsigned int latency_qd_high;
310 unsigned int latency_qd_low;
311 unsigned int latency_failed;
312 uint64_t latency_ios;
313 int latency_end_run;
314
315 /*
316 * read/write mixed workload state
317 */
318 struct frand_state rwmix_state;
319 unsigned long rwmix_issues;
320 enum fio_ddir rwmix_ddir;
321 unsigned int ddir_seq_nr;
322
323 /*
324 * rand/seq mixed workload state
325 */
326 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
327
328 /*
329 * IO history logs for verification. We use a tree for sorting,
330 * if we are overwriting. Otherwise just use a fifo.
331 */
332 struct rb_root io_hist_tree;
333 struct flist_head io_hist_list;
334 unsigned long io_hist_len;
335
336 /*
337 * For IO replaying
338 */
339 struct flist_head io_log_list;
340
341 /*
342 * For tracking/handling discards
343 */
344 struct flist_head trim_list;
345 unsigned long trim_entries;
346
347 struct flist_head next_rand_list;
348
349 /*
350 * for fileservice, how often to switch to a new file
351 */
352 unsigned int file_service_nr;
353 unsigned int file_service_left;
354 struct fio_file *file_service_file;
355
356 unsigned int sync_file_range_nr;
357
358 /*
359 * For generating file sizes
360 */
361 struct frand_state file_size_state;
362
363 /*
364 * Error counts
365 */
366 unsigned int total_err_count;
367 int first_error;
368
369 struct fio_flow *flow;
370
371 /*
372 * Can be overloaded by profiles
373 */
374 struct prof_io_ops prof_io_ops;
375 void *prof_data;
376
377 void *pinned_mem;
378};
379
380/*
381 * when should interactive ETA output be generated
382 */
383enum {
384 FIO_ETA_AUTO,
385 FIO_ETA_ALWAYS,
386 FIO_ETA_NEVER,
387};
388
389#define __td_verror(td, err, msg, func) \
390 do { \
391 unsigned int ____e = (err); \
392 if ((td)->error) \
393 break; \
394 (td)->error = ____e; \
395 if (!(td)->first_error) \
396 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
397 } while (0)
398
399
400#define td_clear_error(td) do { \
401 (td)->error = 0; \
402 if ((td)->parent) \
403 (td)->parent->error = 0; \
404} while (0)
405
406#define td_verror(td, err, func) do { \
407 __td_verror((td), (err), strerror((err)), (func)); \
408 if ((td)->parent) \
409 __td_verror((td)->parent, (err), strerror((err)), (func)); \
410} while (0)
411
412#define td_vmsg(td, err, msg, func) do { \
413 __td_verror((td), (err), (msg), (func)); \
414 if ((td)->parent) \
415 __td_verror((td)->parent, (err), (msg), (func)); \
416} while (0)
417
418#define __fio_stringify_1(x) #x
419#define __fio_stringify(x) __fio_stringify_1(x)
420
421extern int exitall_on_terminate;
422extern unsigned int thread_number;
423extern unsigned int stat_number;
424extern int shm_id;
425extern int groupid;
426extern int output_format;
427extern int append_terse_output;
428extern int temp_stall_ts;
429extern uintptr_t page_mask, page_size;
430extern int read_only;
431extern int eta_print;
432extern int eta_new_line;
433extern unsigned long done_secs;
434extern char *job_section;
435extern int fio_gtod_offload;
436extern int fio_gtod_cpu;
437extern enum fio_cs fio_clock_source;
438extern int fio_clock_source_set;
439extern int warnings_fatal;
440extern int terse_version;
441extern int is_backend;
442extern int nr_clients;
443extern int log_syslog;
444extern int status_interval;
445extern const char fio_version_string[];
446extern int helper_do_stat;
447extern pthread_cond_t helper_cond;
448extern char *trigger_file;
449extern char *trigger_cmd;
450extern char *trigger_remote_cmd;
451extern long long trigger_timeout;
452extern char *aux_path;
453
454extern struct thread_data *threads;
455
456static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
457{
458 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
459}
460
461#define REAL_MAX_JOBS 2048
462
463static inline int should_fsync(struct thread_data *td)
464{
465 if (td->last_was_sync)
466 return 0;
467 if (td_write(td) || td_rw(td) || td->o.override_sync)
468 return 1;
469
470 return 0;
471}
472
473/*
474 * Init/option functions
475 */
476extern int __must_check fio_init_options(void);
477extern int __must_check parse_options(int, char **);
478extern int parse_jobs_ini(char *, int, int, int);
479extern int parse_cmd_line(int, char **, int);
480extern int fio_backend(struct sk_out *);
481extern void reset_fio_state(void);
482extern void clear_io_state(struct thread_data *, int);
483extern int fio_options_parse(struct thread_data *, char **, int);
484extern void fio_keywords_init(void);
485extern void fio_keywords_exit(void);
486extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
487extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
488extern void fio_fill_default_options(struct thread_data *);
489extern int fio_show_option_help(const char *);
490extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
491extern void fio_options_dup_and_init(struct option *);
492extern void fio_options_mem_dupe(struct thread_data *);
493extern void options_mem_dupe(void *data, struct fio_option *options);
494extern void td_fill_rand_seeds(struct thread_data *);
495extern void add_job_opts(const char **, int);
496extern char *num2str(uint64_t, int, int, int, int);
497extern int ioengine_load(struct thread_data *);
498extern int parse_dryrun(void);
499extern int fio_running_or_pending_io_threads(void);
500extern int fio_set_fd_nonblocking(int, const char *);
501extern void sig_show_status(int sig);
502extern struct thread_data *get_global_options(void);
503
504extern uintptr_t page_mask;
505extern uintptr_t page_size;
506extern int initialize_fio(char *envp[]);
507extern void deinitialize_fio(void);
508
509#define FIO_GETOPT_JOB 0x89000000
510#define FIO_GETOPT_IOENGINE 0x98000000
511#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
512
513/*
514 * ETA/status stuff
515 */
516extern void print_thread_status(void);
517extern void print_status_init(int);
518extern char *fio_uint_to_kmg(unsigned int val);
519
520/*
521 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
522 * will never back again. It may cycle between running/verififying/fsyncing.
523 * Once the thread reaches TD_EXITED, it is just waiting for the core to
524 * reap it.
525 */
526enum {
527 TD_NOT_CREATED = 0,
528 TD_CREATED,
529 TD_INITIALIZED,
530 TD_RAMP,
531 TD_SETTING_UP,
532 TD_RUNNING,
533 TD_PRE_READING,
534 TD_VERIFYING,
535 TD_FSYNCING,
536 TD_FINISHING,
537 TD_EXITED,
538 TD_REAPED,
539 TD_LAST,
540};
541
542extern void td_set_runstate(struct thread_data *, int);
543extern int td_bump_runstate(struct thread_data *, int);
544extern void td_restore_runstate(struct thread_data *, int);
545extern const char *runstate_to_name(int runstate);
546
547/*
548 * Allow 60 seconds for a job to quit on its own, otherwise reap with
549 * a vengeance.
550 */
551#define FIO_REAP_TIMEOUT 60
552
553#define TERMINATE_ALL (-1U)
554extern void fio_terminate_threads(unsigned int);
555extern void fio_mark_td_terminate(struct thread_data *);
556
557/*
558 * Memory helpers
559 */
560extern int __must_check fio_pin_memory(struct thread_data *);
561extern void fio_unpin_memory(struct thread_data *);
562extern int __must_check allocate_io_mem(struct thread_data *);
563extern void free_io_mem(struct thread_data *);
564extern void free_threads_shm(void);
565
566/*
567 * Reset stats after ramp time completes
568 */
569extern void reset_all_stats(struct thread_data *);
570
571/*
572 * blktrace support
573 */
574#ifdef FIO_HAVE_BLKTRACE
575extern int is_blktrace(const char *, int *);
576extern int load_blktrace(struct thread_data *, const char *, int);
577#endif
578
579extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
580 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
581 struct timeval *comp_time);
582
583/*
584 * Latency target helpers
585 */
586extern void lat_target_check(struct thread_data *);
587extern void lat_target_init(struct thread_data *);
588extern void lat_target_reset(struct thread_data *);
589
590#define for_each_td(td, i) \
591 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
592#define for_each_file(td, f, i) \
593 if ((td)->files_index) \
594 for ((i) = 0, (f) = (td)->files[0]; \
595 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
596 (i)++)
597
598#define fio_assert(td, cond) do { \
599 if (!(cond)) { \
600 int *__foo = NULL; \
601 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
602 td_set_runstate((td), TD_EXITED); \
603 (td)->error = EFAULT; \
604 *__foo = 0; \
605 } \
606} while (0)
607
608static inline int fio_fill_issue_time(struct thread_data *td)
609{
610 if (td->o.read_iolog_file ||
611 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
612 return 1;
613
614 return 0;
615}
616
617static inline int __should_check_rate(struct thread_data *td,
618 enum fio_ddir ddir)
619{
620 struct thread_options *o = &td->o;
621
622 /*
623 * If some rate setting was given, we need to check it
624 */
625 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
626 o->rate_iops_min[ddir])
627 return 1;
628
629 return 0;
630}
631
632static inline int should_check_rate(struct thread_data *td)
633{
634 int ret = 0;
635
636 if (td->bytes_done[DDIR_READ])
637 ret |= __should_check_rate(td, DDIR_READ);
638 if (td->bytes_done[DDIR_WRITE])
639 ret |= __should_check_rate(td, DDIR_WRITE);
640 if (td->bytes_done[DDIR_TRIM])
641 ret |= __should_check_rate(td, DDIR_TRIM);
642
643 return ret;
644}
645
646static inline unsigned int td_max_bs(struct thread_data *td)
647{
648 unsigned int max_bs;
649
650 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
651 return max(td->o.max_bs[DDIR_TRIM], max_bs);
652}
653
654static inline unsigned int td_min_bs(struct thread_data *td)
655{
656 unsigned int min_bs;
657
658 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
659 return min(td->o.min_bs[DDIR_TRIM], min_bs);
660}
661
662static inline int td_async_processing(struct thread_data *td)
663{
664 return (td->flags & TD_F_NEED_LOCK) != 0;
665}
666
667/*
668 * We currently only need to do locking if we have verifier threads
669 * accessing our internal structures too
670 */
671static inline void td_io_u_lock(struct thread_data *td)
672{
673 if (td_async_processing(td))
674 pthread_mutex_lock(&td->io_u_lock);
675}
676
677static inline void td_io_u_unlock(struct thread_data *td)
678{
679 if (td_async_processing(td))
680 pthread_mutex_unlock(&td->io_u_lock);
681}
682
683static inline void td_io_u_free_notify(struct thread_data *td)
684{
685 if (td_async_processing(td))
686 pthread_cond_signal(&td->free_cond);
687}
688
689extern const char *fio_get_arch_string(int);
690extern const char *fio_get_os_string(int);
691
692#ifdef FIO_INTERNAL
693#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
694#define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
695#endif
696
697enum {
698 __FIO_OUTPUT_TERSE = 0,
699 __FIO_OUTPUT_JSON = 1,
700 __FIO_OUTPUT_NORMAL = 2,
701 __FIO_OUTPUT_JSON_PLUS = 3,
702 FIO_OUTPUT_NR = 4,
703
704 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
705 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
706 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
707 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
708};
709
710enum {
711 FIO_RAND_DIST_RANDOM = 0,
712 FIO_RAND_DIST_ZIPF,
713 FIO_RAND_DIST_PARETO,
714 FIO_RAND_DIST_GAUSS,
715};
716
717#define FIO_DEF_ZIPF 1.1
718#define FIO_DEF_PARETO 0.2
719
720enum {
721 FIO_RAND_GEN_TAUSWORTHE = 0,
722 FIO_RAND_GEN_LFSR,
723 FIO_RAND_GEN_TAUSWORTHE64,
724};
725
726enum {
727 FIO_CPUS_SHARED = 0,
728 FIO_CPUS_SPLIT,
729};
730
731extern void exec_trigger(const char *);
732extern void check_trigger_file(void);
733
734#endif
fio - Flexible IO Tester