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