projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Improve precision of the io_limit setting
[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
44#ifdef CONFIG_SOLARISAIO
45#include <sys/asynch.h>
46#endif
47
48#ifdef CONFIG_LIBNUMA
49#include <linux/mempolicy.h>
50#include <numa.h>
51
52/*
53 * "local" is pseudo-policy
54 */
55#define MPOL_LOCAL MPOL_MAX
56#endif
57
58/*
59 * offset generator types
60 */
61enum {
62 RW_SEQ_SEQ = 0,
63 RW_SEQ_IDENT,
64};
65
66enum {
67 TD_F_VER_BACKLOG = 1,
68 TD_F_TRIM_BACKLOG = 2,
69 TD_F_READ_IOLOG = 4,
70 TD_F_REFILL_BUFFERS = 8,
71 TD_F_SCRAMBLE_BUFFERS = 16,
72 TD_F_VER_NONE = 32,
73 TD_F_PROFILE_OPS = 64,
74 TD_F_COMPRESS = 128,
75 TD_F_NOIO = 256,
76 TD_F_COMPRESS_LOG = 512,
77 TD_F_VSTATE_SAVED = 1024,
78};
79
80enum {
81 FIO_RAND_BS_OFF = 0,
82 FIO_RAND_VER_OFF,
83 FIO_RAND_MIX_OFF,
84 FIO_RAND_FILE_OFF,
85 FIO_RAND_BLOCK_OFF,
86 FIO_RAND_FILE_SIZE_OFF,
87 FIO_RAND_TRIM_OFF,
88 FIO_RAND_BUF_OFF,
89 FIO_RAND_SEQ_RAND_READ_OFF,
90 FIO_RAND_SEQ_RAND_WRITE_OFF,
91 FIO_RAND_SEQ_RAND_TRIM_OFF,
92 FIO_RAND_START_DELAY,
93 FIO_DEDUPE_OFF,
94 FIO_RAND_NR_OFFS,
95};
96
97/*
98 * This describes a single thread/process executing a fio job.
99 */
100struct thread_data {
101 struct thread_options o;
102 unsigned long flags;
103 void *eo;
104 char verror[FIO_VERROR_SIZE];
105 pthread_t thread;
106 unsigned int thread_number;
107 unsigned int subjob_number;
108 unsigned int groupid;
109 struct thread_stat ts;
110
111 int client_type;
112
113 struct io_log *slat_log;
114 struct io_log *clat_log;
115 struct io_log *lat_log;
116 struct io_log *bw_log;
117 struct io_log *iops_log;
118
119 struct tp_data *tp_data;
120
121 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
122 struct timeval bw_sample_time;
123
124 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
125 struct timeval iops_sample_time;
126
127 /*
128 * Tracks the last iodepth number of completed writes, if data
129 * verification is enabled
130 */
131 uint64_t *last_write_comp;
132 unsigned int last_write_idx;
133
134 volatile int update_rusage;
135 struct fio_mutex *rusage_sem;
136 struct rusage ru_start;
137 struct rusage ru_end;
138
139 struct fio_file **files;
140 unsigned char *file_locks;
141 unsigned int files_size;
142 unsigned int files_index;
143 unsigned int nr_open_files;
144 unsigned int nr_done_files;
145 unsigned int nr_normal_files;
146 union {
147 unsigned int next_file;
148 struct frand_state next_file_state;
149 };
150 int error;
151 int sig;
152 int done;
153 int stop_io;
154 pid_t pid;
155 char *orig_buffer;
156 size_t orig_buffer_size;
157 volatile int terminate;
158 volatile int runstate;
159 unsigned int last_was_sync;
160 enum fio_ddir last_ddir;
161
162 int mmapfd;
163
164 void *iolog_buf;
165 FILE *iolog_f;
166
167 char *sysfs_root;
168
169 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
170
171 struct frand_state bsrange_state;
172 struct frand_state verify_state;
173 struct frand_state trim_state;
174 struct frand_state delay_state;
175
176 struct frand_state buf_state;
177 struct frand_state buf_state_prev;
178 struct frand_state dedupe_state;
179
180 unsigned int verify_batch;
181 unsigned int trim_batch;
182
183 struct thread_io_list *vstate;
184
185 int shm_id;
186
187 /*
188 * IO engine hooks, contains everything needed to submit an io_u
189 * to any of the available IO engines.
190 */
191 struct ioengine_ops *io_ops;
192
193 /*
194 * Queue depth of io_u's that fio MIGHT do
195 */
196 unsigned int cur_depth;
197
198 /*
199 * io_u's about to be committed
200 */
201 unsigned int io_u_queued;
202
203 /*
204 * io_u's submitted but not completed yet
205 */
206 unsigned int io_u_in_flight;
207
208 /*
209 * List of free and busy io_u's
210 */
211 struct io_u_ring io_u_requeues;
212 struct io_u_queue io_u_freelist;
213 struct io_u_queue io_u_all;
214 pthread_mutex_t io_u_lock;
215 pthread_cond_t free_cond;
216
217 /*
218 * async verify offload
219 */
220 struct flist_head verify_list;
221 pthread_t *verify_threads;
222 unsigned int nr_verify_threads;
223 pthread_cond_t verify_cond;
224 int verify_thread_exit;
225
226 /*
227 * Rate state
228 */
229 uint64_t rate_bps[DDIR_RWDIR_CNT];
230 long rate_pending_usleep[DDIR_RWDIR_CNT];
231 unsigned long rate_bytes[DDIR_RWDIR_CNT];
232 unsigned long rate_blocks[DDIR_RWDIR_CNT];
233 struct timeval lastrate[DDIR_RWDIR_CNT];
234
235 uint64_t total_io_size;
236 uint64_t fill_device_size;
237
238 /*
239 * Issue side
240 */
241 uint64_t io_issues[DDIR_RWDIR_CNT];
242 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
243
244 /*
245 * Completions
246 */
247 uint64_t io_blocks[DDIR_RWDIR_CNT];
248 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
249 uint64_t io_bytes[DDIR_RWDIR_CNT];
250 uint64_t io_skip_bytes;
251 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
252 uint64_t zone_bytes;
253 struct fio_mutex *mutex;
254
255 /*
256 * State for random io, a bitmap of blocks done vs not done
257 */
258 struct frand_state random_state;
259
260 struct timeval start; /* start of this loop */
261 struct timeval epoch; /* time job was started */
262 struct timeval last_issue;
263 long time_offset;
264 struct timeval tv_cache;
265 struct timeval terminate_time;
266 unsigned int tv_cache_nr;
267 unsigned int tv_cache_mask;
268 unsigned int ramp_time_over;
269
270 /*
271 * Time since last latency_window was started
272 */
273 struct timeval latency_ts;
274 unsigned int latency_qd;
275 unsigned int latency_qd_high;
276 unsigned int latency_qd_low;
277 unsigned int latency_failed;
278 uint64_t latency_ios;
279 int latency_end_run;
280
281 /*
282 * read/write mixed workload state
283 */
284 struct frand_state rwmix_state;
285 unsigned long rwmix_issues;
286 enum fio_ddir rwmix_ddir;
287 unsigned int ddir_seq_nr;
288
289 /*
290 * rand/seq mixed workload state
291 */
292 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
293
294 /*
295 * IO history logs for verification. We use a tree for sorting,
296 * if we are overwriting. Otherwise just use a fifo.
297 */
298 struct rb_root io_hist_tree;
299 struct flist_head io_hist_list;
300 unsigned long io_hist_len;
301
302 /*
303 * For IO replaying
304 */
305 struct flist_head io_log_list;
306
307 /*
308 * For tracking/handling discards
309 */
310 struct flist_head trim_list;
311 unsigned long trim_entries;
312
313 struct flist_head next_rand_list;
314
315 /*
316 * for fileservice, how often to switch to a new file
317 */
318 unsigned int file_service_nr;
319 unsigned int file_service_left;
320 struct fio_file *file_service_file;
321
322 unsigned int sync_file_range_nr;
323
324 /*
325 * For generating file sizes
326 */
327 struct frand_state file_size_state;
328
329 /*
330 * Error counts
331 */
332 unsigned int total_err_count;
333 int first_error;
334
335 struct fio_flow *flow;
336
337 /*
338 * Can be overloaded by profiles
339 */
340 struct prof_io_ops prof_io_ops;
341 void *prof_data;
342
343 void *pinned_mem;
344};
345
346/*
347 * when should interactive ETA output be generated
348 */
349enum {
350 FIO_ETA_AUTO,
351 FIO_ETA_ALWAYS,
352 FIO_ETA_NEVER,
353};
354
355#define __td_verror(td, err, msg, func) \
356 do { \
357 unsigned int ____e = (err); \
358 if ((td)->error) \
359 break; \
360 (td)->error = ____e; \
361 if (!(td)->first_error) \
362 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
363 } while (0)
364
365
366#define td_clear_error(td) \
367 (td)->error = 0;
368#define td_verror(td, err, func) \
369 __td_verror((td), (err), strerror((err)), (func))
370#define td_vmsg(td, err, msg, func) \
371 __td_verror((td), (err), (msg), (func))
372
373#define __fio_stringify_1(x) #x
374#define __fio_stringify(x) __fio_stringify_1(x)
375
376extern int exitall_on_terminate;
377extern unsigned int thread_number;
378extern unsigned int stat_number;
379extern int shm_id;
380extern int groupid;
381extern int output_format;
382extern int append_terse_output;
383extern int temp_stall_ts;
384extern uintptr_t page_mask, page_size;
385extern int read_only;
386extern int eta_print;
387extern int eta_new_line;
388extern unsigned long done_secs;
389extern char *job_section;
390extern int fio_gtod_offload;
391extern int fio_gtod_cpu;
392extern enum fio_cs fio_clock_source;
393extern int fio_clock_source_set;
394extern int warnings_fatal;
395extern int terse_version;
396extern int is_backend;
397extern int nr_clients;
398extern int log_syslog;
399extern int status_interval;
400extern const char fio_version_string[];
401extern int helper_do_stat;
402extern pthread_cond_t helper_cond;
403extern char *trigger_file;
404extern char *trigger_cmd;
405extern char *trigger_remote_cmd;
406extern long long trigger_timeout;
407
408extern struct thread_data *threads;
409
410static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
411{
412 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
413}
414
415#define REAL_MAX_JOBS 2048
416
417static inline int should_fsync(struct thread_data *td)
418{
419 if (td->last_was_sync)
420 return 0;
421 if (td_write(td) || td_rw(td) || td->o.override_sync)
422 return 1;
423
424 return 0;
425}
426
427/*
428 * Init/option functions
429 */
430extern int __must_check fio_init_options(void);
431extern int __must_check parse_options(int, char **);
432extern int parse_jobs_ini(char *, int, int, int);
433extern int parse_cmd_line(int, char **, int);
434extern int fio_backend(void);
435extern void reset_fio_state(void);
436extern void clear_io_state(struct thread_data *);
437extern int fio_options_parse(struct thread_data *, char **, int, int);
438extern void fio_keywords_init(void);
439extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
440extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
441extern void fio_fill_default_options(struct thread_data *);
442extern int fio_show_option_help(const char *);
443extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
444extern void fio_options_dup_and_init(struct option *);
445extern void fio_options_mem_dupe(struct thread_data *);
446extern void options_mem_dupe(void *data, struct fio_option *options);
447extern void td_fill_rand_seeds(struct thread_data *);
448extern void add_job_opts(const char **, int);
449extern char *num2str(uint64_t, int, int, int, int);
450extern int ioengine_load(struct thread_data *);
451extern int parse_dryrun(void);
452extern int fio_running_or_pending_io_threads(void);
453extern int fio_set_fd_nonblocking(int, const char *);
454
455extern uintptr_t page_mask;
456extern uintptr_t page_size;
457extern int initialize_fio(char *envp[]);
458
459#define FIO_GETOPT_JOB 0x89000000
460#define FIO_GETOPT_IOENGINE 0x98000000
461#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
462
463/*
464 * ETA/status stuff
465 */
466extern void print_thread_status(void);
467extern void print_status_init(int);
468extern char *fio_uint_to_kmg(unsigned int val);
469
470/*
471 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
472 * will never back again. It may cycle between running/verififying/fsyncing.
473 * Once the thread reaches TD_EXITED, it is just waiting for the core to
474 * reap it.
475 */
476enum {
477 TD_NOT_CREATED = 0,
478 TD_CREATED,
479 TD_INITIALIZED,
480 TD_RAMP,
481 TD_SETTING_UP,
482 TD_RUNNING,
483 TD_PRE_READING,
484 TD_VERIFYING,
485 TD_FSYNCING,
486 TD_FINISHING,
487 TD_EXITED,
488 TD_REAPED,
489};
490
491extern void td_set_runstate(struct thread_data *, int);
492extern int td_bump_runstate(struct thread_data *, int);
493extern void td_restore_runstate(struct thread_data *, int);
494
495/*
496 * Allow 60 seconds for a job to quit on its own, otherwise reap with
497 * a vengeance.
498 */
499#define FIO_REAP_TIMEOUT 60
500
501#define TERMINATE_ALL (-1)
502extern void fio_terminate_threads(int);
503extern void fio_mark_td_terminate(struct thread_data *);
504
505/*
506 * Memory helpers
507 */
508extern int __must_check fio_pin_memory(struct thread_data *);
509extern void fio_unpin_memory(struct thread_data *);
510extern int __must_check allocate_io_mem(struct thread_data *);
511extern void free_io_mem(struct thread_data *);
512extern void free_threads_shm(void);
513
514/*
515 * Reset stats after ramp time completes
516 */
517extern void reset_all_stats(struct thread_data *);
518
519/*
520 * blktrace support
521 */
522#ifdef FIO_HAVE_BLKTRACE
523extern int is_blktrace(const char *, int *);
524extern int load_blktrace(struct thread_data *, const char *, int);
525#endif
526
527/*
528 * Latency target helpers
529 */
530extern void lat_target_check(struct thread_data *);
531extern void lat_target_init(struct thread_data *);
532extern void lat_target_reset(struct thread_data *);
533
534#define for_each_td(td, i) \
535 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
536#define for_each_file(td, f, i) \
537 if ((td)->files_index) \
538 for ((i) = 0, (f) = (td)->files[0]; \
539 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
540 (i)++)
541
542#define fio_assert(td, cond) do { \
543 if (!(cond)) { \
544 int *__foo = NULL; \
545 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
546 td_set_runstate((td), TD_EXITED); \
547 (td)->error = EFAULT; \
548 *__foo = 0; \
549 } \
550} while (0)
551
552static inline int fio_fill_issue_time(struct thread_data *td)
553{
554 if (td->o.read_iolog_file ||
555 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
556 return 1;
557
558 return 0;
559}
560
561static inline int __should_check_rate(struct thread_data *td,
562 enum fio_ddir ddir)
563{
564 struct thread_options *o = &td->o;
565
566 /*
567 * If some rate setting was given, we need to check it
568 */
569 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
570 o->rate_iops_min[ddir])
571 return 1;
572
573 return 0;
574}
575
576static inline int should_check_rate(struct thread_data *td,
577 uint64_t *bytes_done)
578{
579 int ret = 0;
580
581 if (bytes_done[DDIR_READ])
582 ret |= __should_check_rate(td, DDIR_READ);
583 if (bytes_done[DDIR_WRITE])
584 ret |= __should_check_rate(td, DDIR_WRITE);
585 if (bytes_done[DDIR_TRIM])
586 ret |= __should_check_rate(td, DDIR_TRIM);
587
588 return ret;
589}
590
591static inline unsigned int td_max_bs(struct thread_data *td)
592{
593 unsigned int max_bs;
594
595 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
596 return max(td->o.max_bs[DDIR_TRIM], max_bs);
597}
598
599static inline unsigned int td_min_bs(struct thread_data *td)
600{
601 unsigned int min_bs;
602
603 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
604 return min(td->o.min_bs[DDIR_TRIM], min_bs);
605}
606
607static inline int is_power_of_2(uint64_t val)
608{
609 return (val != 0 && ((val & (val - 1)) == 0));
610}
611
612/*
613 * We currently only need to do locking if we have verifier threads
614 * accessing our internal structures too
615 */
616static inline void td_io_u_lock(struct thread_data *td)
617{
618 if (td->o.verify_async)
619 pthread_mutex_lock(&td->io_u_lock);
620}
621
622static inline void td_io_u_unlock(struct thread_data *td)
623{
624 if (td->o.verify_async)
625 pthread_mutex_unlock(&td->io_u_lock);
626}
627
628static inline void td_io_u_free_notify(struct thread_data *td)
629{
630 if (td->o.verify_async)
631 pthread_cond_signal(&td->free_cond);
632}
633
634extern const char *fio_get_arch_string(int);
635extern const char *fio_get_os_string(int);
636
637#ifdef FIO_INTERNAL
638#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
639#endif
640
641enum {
642 FIO_OUTPUT_TERSE = 0,
643 FIO_OUTPUT_JSON,
644 FIO_OUTPUT_NORMAL,
645};
646
647enum {
648 FIO_RAND_DIST_RANDOM = 0,
649 FIO_RAND_DIST_ZIPF,
650 FIO_RAND_DIST_PARETO,
651};
652
653#define FIO_DEF_ZIPF 1.1
654#define FIO_DEF_PARETO 0.2
655
656enum {
657 FIO_RAND_GEN_TAUSWORTHE = 0,
658 FIO_RAND_GEN_LFSR,
659};
660
661enum {
662 FIO_CPUS_SHARED = 0,
663 FIO_CPUS_SPLIT,
664};
665
666extern void exec_trigger(const char *);
667extern void check_trigger_file(void);
668
669#endif
fio - Flexible IO Tester