projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
This commit / feature adds completion latency histogram output to fio, piggybacking
[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 TD_F_REGROW_LOGS = 1U << 14,
83};
84
85enum {
86 FIO_RAND_BS_OFF = 0,
87 FIO_RAND_VER_OFF,
88 FIO_RAND_MIX_OFF,
89 FIO_RAND_FILE_OFF,
90 FIO_RAND_BLOCK_OFF,
91 FIO_RAND_FILE_SIZE_OFF,
92 FIO_RAND_TRIM_OFF,
93 FIO_RAND_BUF_OFF,
94 FIO_RAND_SEQ_RAND_READ_OFF,
95 FIO_RAND_SEQ_RAND_WRITE_OFF,
96 FIO_RAND_SEQ_RAND_TRIM_OFF,
97 FIO_RAND_START_DELAY,
98 FIO_DEDUPE_OFF,
99 FIO_RAND_POISSON_OFF,
100 FIO_RAND_ZONE_OFF,
101 FIO_RAND_NR_OFFS,
102};
103
104enum {
105 IO_MODE_INLINE = 0,
106 IO_MODE_OFFLOAD = 1,
107
108 RATE_PROCESS_LINEAR = 0,
109 RATE_PROCESS_POISSON = 1,
110};
111
112/*
113 * Per-thread/process specific data. Only used for the network client
114 * for now.
115 */
116struct sk_out;
117void sk_out_assign(struct sk_out *);
118void sk_out_drop(void);
119
120struct zone_split_index {
121 uint8_t size_perc;
122 uint8_t size_perc_prev;
123};
124
125/*
126 * This describes a single thread/process executing a fio job.
127 */
128struct thread_data {
129 struct thread_options o;
130 struct flist_head opt_list;
131 unsigned long flags;
132 void *eo;
133 char verror[FIO_VERROR_SIZE];
134 pthread_t thread;
135 unsigned int thread_number;
136 unsigned int subjob_number;
137 unsigned int groupid;
138 struct thread_stat ts;
139
140 int client_type;
141
142 struct io_log *slat_log;
143 struct io_log *clat_log;
144 struct io_log *clat_hist_log;
145 struct io_log *lat_log;
146 struct io_log *bw_log;
147 struct io_log *iops_log;
148
149 struct workqueue log_compress_wq;
150
151 struct thread_data *parent;
152
153 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
154 struct timeval bw_sample_time;
155
156 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
157 struct timeval iops_sample_time;
158
159 volatile int update_rusage;
160 struct fio_mutex *rusage_sem;
161 struct rusage ru_start;
162 struct rusage ru_end;
163
164 struct fio_file **files;
165 unsigned char *file_locks;
166 unsigned int files_size;
167 unsigned int files_index;
168 unsigned int nr_open_files;
169 unsigned int nr_done_files;
170 unsigned int nr_normal_files;
171 union {
172 unsigned int next_file;
173 struct frand_state next_file_state;
174 };
175 union {
176 struct zipf_state next_file_zipf;
177 struct gauss_state next_file_gauss;
178 };
179 union {
180 double zipf_theta;
181 double pareto_h;
182 double gauss_dev;
183 };
184 int error;
185 int sig;
186 int done;
187 int stop_io;
188 pid_t pid;
189 char *orig_buffer;
190 size_t orig_buffer_size;
191 volatile int terminate;
192 volatile int runstate;
193 unsigned int last_was_sync;
194 enum fio_ddir last_ddir;
195
196 int mmapfd;
197
198 void *iolog_buf;
199 FILE *iolog_f;
200
201 char *sysfs_root;
202
203 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
204
205 struct frand_state bsrange_state;
206 struct frand_state verify_state;
207 struct frand_state trim_state;
208 struct frand_state delay_state;
209
210 struct frand_state buf_state;
211 struct frand_state buf_state_prev;
212 struct frand_state dedupe_state;
213 struct frand_state zone_state;
214
215 struct zone_split_index **zone_state_index;
216
217 unsigned int verify_batch;
218 unsigned int trim_batch;
219
220 struct thread_io_list *vstate;
221
222 int shm_id;
223
224 /*
225 * IO engine hooks, contains everything needed to submit an io_u
226 * to any of the available IO engines.
227 */
228 struct ioengine_ops *io_ops;
229
230 /*
231 * IO engine private data and dlhandle.
232 */
233 void *io_ops_data;
234 void *io_ops_dlhandle;
235
236 /*
237 * Queue depth of io_u's that fio MIGHT do
238 */
239 unsigned int cur_depth;
240
241 /*
242 * io_u's about to be committed
243 */
244 unsigned int io_u_queued;
245
246 /*
247 * io_u's submitted but not completed yet
248 */
249 unsigned int io_u_in_flight;
250
251 /*
252 * List of free and busy io_u's
253 */
254 struct io_u_ring io_u_requeues;
255 struct io_u_queue io_u_freelist;
256 struct io_u_queue io_u_all;
257 pthread_mutex_t io_u_lock;
258 pthread_cond_t free_cond;
259
260 /*
261 * async verify offload
262 */
263 struct flist_head verify_list;
264 pthread_t *verify_threads;
265 unsigned int nr_verify_threads;
266 pthread_cond_t verify_cond;
267 int verify_thread_exit;
268
269 /*
270 * Rate state
271 */
272 uint64_t rate_bps[DDIR_RWDIR_CNT];
273 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
274 unsigned long rate_bytes[DDIR_RWDIR_CNT];
275 unsigned long rate_blocks[DDIR_RWDIR_CNT];
276 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
277 struct timeval lastrate[DDIR_RWDIR_CNT];
278 int64_t last_usec;
279 struct frand_state poisson_state;
280
281 /*
282 * Enforced rate submission/completion workqueue
283 */
284 struct workqueue io_wq;
285
286 uint64_t total_io_size;
287 uint64_t fill_device_size;
288
289 /*
290 * Issue side
291 */
292 uint64_t io_issues[DDIR_RWDIR_CNT];
293 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
294 uint64_t loops;
295
296 /*
297 * Completions
298 */
299 uint64_t io_blocks[DDIR_RWDIR_CNT];
300 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
301 uint64_t io_bytes[DDIR_RWDIR_CNT];
302 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
303 uint64_t io_skip_bytes;
304 uint64_t zone_bytes;
305 struct fio_mutex *mutex;
306 uint64_t bytes_done[DDIR_RWDIR_CNT];
307
308 /*
309 * State for random io, a bitmap of blocks done vs not done
310 */
311 struct frand_state random_state;
312
313 struct timeval start; /* start of this loop */
314 struct timeval epoch; /* time job was started */
315 struct timeval last_issue;
316 long time_offset;
317 struct timeval tv_cache;
318 struct timeval terminate_time;
319 unsigned int tv_cache_nr;
320 unsigned int tv_cache_mask;
321 unsigned int ramp_time_over;
322
323 /*
324 * Time since last latency_window was started
325 */
326 struct timeval latency_ts;
327 unsigned int latency_qd;
328 unsigned int latency_qd_high;
329 unsigned int latency_qd_low;
330 unsigned int latency_failed;
331 uint64_t latency_ios;
332 int latency_end_run;
333
334 /*
335 * read/write mixed workload state
336 */
337 struct frand_state rwmix_state;
338 unsigned long rwmix_issues;
339 enum fio_ddir rwmix_ddir;
340 unsigned int ddir_seq_nr;
341
342 /*
343 * rand/seq mixed workload state
344 */
345 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
346
347 /*
348 * IO history logs for verification. We use a tree for sorting,
349 * if we are overwriting. Otherwise just use a fifo.
350 */
351 struct rb_root io_hist_tree;
352 struct flist_head io_hist_list;
353 unsigned long io_hist_len;
354
355 /*
356 * For IO replaying
357 */
358 struct flist_head io_log_list;
359
360 /*
361 * For tracking/handling discards
362 */
363 struct flist_head trim_list;
364 unsigned long trim_entries;
365
366 struct flist_head next_rand_list;
367
368 /*
369 * for fileservice, how often to switch to a new file
370 */
371 unsigned int file_service_nr;
372 unsigned int file_service_left;
373 struct fio_file *file_service_file;
374
375 unsigned int sync_file_range_nr;
376
377 /*
378 * For generating file sizes
379 */
380 struct frand_state file_size_state;
381
382 /*
383 * Error counts
384 */
385 unsigned int total_err_count;
386 int first_error;
387
388 struct fio_flow *flow;
389
390 /*
391 * Can be overloaded by profiles
392 */
393 struct prof_io_ops prof_io_ops;
394 void *prof_data;
395
396 void *pinned_mem;
397};
398
399/*
400 * when should interactive ETA output be generated
401 */
402enum {
403 FIO_ETA_AUTO,
404 FIO_ETA_ALWAYS,
405 FIO_ETA_NEVER,
406};
407
408#define __td_verror(td, err, msg, func) \
409 do { \
410 unsigned int ____e = (err); \
411 if ((td)->error) \
412 break; \
413 (td)->error = ____e; \
414 if (!(td)->first_error) \
415 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
416 } while (0)
417
418
419#define td_clear_error(td) do { \
420 (td)->error = 0; \
421 if ((td)->parent) \
422 (td)->parent->error = 0; \
423} while (0)
424
425#define td_verror(td, err, func) do { \
426 __td_verror((td), (err), strerror((err)), (func)); \
427 if ((td)->parent) \
428 __td_verror((td)->parent, (err), strerror((err)), (func)); \
429} while (0)
430
431#define td_vmsg(td, err, msg, func) do { \
432 __td_verror((td), (err), (msg), (func)); \
433 if ((td)->parent) \
434 __td_verror((td)->parent, (err), (msg), (func)); \
435} while (0)
436
437#define __fio_stringify_1(x) #x
438#define __fio_stringify(x) __fio_stringify_1(x)
439
440extern int exitall_on_terminate;
441extern unsigned int thread_number;
442extern unsigned int stat_number;
443extern int shm_id;
444extern int groupid;
445extern int output_format;
446extern int append_terse_output;
447extern int temp_stall_ts;
448extern uintptr_t page_mask, page_size;
449extern int read_only;
450extern int eta_print;
451extern int eta_new_line;
452extern unsigned long done_secs;
453extern char *job_section;
454extern int fio_gtod_offload;
455extern int fio_gtod_cpu;
456extern enum fio_cs fio_clock_source;
457extern int fio_clock_source_set;
458extern int warnings_fatal;
459extern int terse_version;
460extern int is_backend;
461extern int nr_clients;
462extern int log_syslog;
463extern int status_interval;
464extern const char fio_version_string[];
465extern char *trigger_file;
466extern char *trigger_cmd;
467extern char *trigger_remote_cmd;
468extern long long trigger_timeout;
469extern char *aux_path;
470
471extern struct thread_data *threads;
472
473static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
474{
475 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
476}
477
478#define REAL_MAX_JOBS 2048
479
480static inline int should_fsync(struct thread_data *td)
481{
482 if (td->last_was_sync)
483 return 0;
484 if (td_write(td) || td_rw(td) || td->o.override_sync)
485 return 1;
486
487 return 0;
488}
489
490/*
491 * Init/option functions
492 */
493extern int __must_check fio_init_options(void);
494extern int __must_check parse_options(int, char **);
495extern int parse_jobs_ini(char *, int, int, int);
496extern int parse_cmd_line(int, char **, int);
497extern int fio_backend(struct sk_out *);
498extern void reset_fio_state(void);
499extern void clear_io_state(struct thread_data *, int);
500extern int fio_options_parse(struct thread_data *, char **, int);
501extern void fio_keywords_init(void);
502extern void fio_keywords_exit(void);
503extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
504extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
505extern void fio_fill_default_options(struct thread_data *);
506extern int fio_show_option_help(const char *);
507extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
508extern void fio_options_dup_and_init(struct option *);
509extern void fio_options_mem_dupe(struct thread_data *);
510extern void options_mem_dupe(void *data, struct fio_option *options);
511extern void td_fill_rand_seeds(struct thread_data *);
512extern void td_fill_verify_state_seed(struct thread_data *);
513extern void add_job_opts(const char **, int);
514extern char *num2str(uint64_t, int, int, int, int);
515extern int ioengine_load(struct thread_data *);
516extern int parse_dryrun(void);
517extern int fio_running_or_pending_io_threads(void);
518extern int fio_set_fd_nonblocking(int, const char *);
519extern void sig_show_status(int sig);
520extern struct thread_data *get_global_options(void);
521
522extern uintptr_t page_mask;
523extern uintptr_t page_size;
524extern int initialize_fio(char *envp[]);
525extern void deinitialize_fio(void);
526
527#define FIO_GETOPT_JOB 0x89000000
528#define FIO_GETOPT_IOENGINE 0x98000000
529#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
530
531/*
532 * ETA/status stuff
533 */
534extern void print_thread_status(void);
535extern void print_status_init(int);
536extern char *fio_uint_to_kmg(unsigned int val);
537
538/*
539 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
540 * will never back again. It may cycle between running/verififying/fsyncing.
541 * Once the thread reaches TD_EXITED, it is just waiting for the core to
542 * reap it.
543 */
544enum {
545 TD_NOT_CREATED = 0,
546 TD_CREATED,
547 TD_INITIALIZED,
548 TD_RAMP,
549 TD_SETTING_UP,
550 TD_RUNNING,
551 TD_PRE_READING,
552 TD_VERIFYING,
553 TD_FSYNCING,
554 TD_FINISHING,
555 TD_EXITED,
556 TD_REAPED,
557 TD_LAST,
558};
559
560extern void td_set_runstate(struct thread_data *, int);
561extern int td_bump_runstate(struct thread_data *, int);
562extern void td_restore_runstate(struct thread_data *, int);
563extern const char *runstate_to_name(int runstate);
564
565/*
566 * Allow 60 seconds for a job to quit on its own, otherwise reap with
567 * a vengeance.
568 */
569#define FIO_REAP_TIMEOUT 60
570
571#define TERMINATE_ALL (-1U)
572extern void fio_terminate_threads(unsigned int);
573extern void fio_mark_td_terminate(struct thread_data *);
574
575/*
576 * Memory helpers
577 */
578extern int __must_check fio_pin_memory(struct thread_data *);
579extern void fio_unpin_memory(struct thread_data *);
580extern int __must_check allocate_io_mem(struct thread_data *);
581extern void free_io_mem(struct thread_data *);
582extern void free_threads_shm(void);
583
584/*
585 * Reset stats after ramp time completes
586 */
587extern void reset_all_stats(struct thread_data *);
588
589/*
590 * blktrace support
591 */
592#ifdef FIO_HAVE_BLKTRACE
593extern int is_blktrace(const char *, int *);
594extern int load_blktrace(struct thread_data *, const char *, int);
595#endif
596
597extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
598 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
599 struct timeval *comp_time);
600
601/*
602 * Latency target helpers
603 */
604extern void lat_target_check(struct thread_data *);
605extern void lat_target_init(struct thread_data *);
606extern void lat_target_reset(struct thread_data *);
607
608#define for_each_td(td, i) \
609 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
610#define for_each_file(td, f, i) \
611 if ((td)->files_index) \
612 for ((i) = 0, (f) = (td)->files[0]; \
613 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
614 (i)++)
615
616#define fio_assert(td, cond) do { \
617 if (!(cond)) { \
618 int *__foo = NULL; \
619 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
620 td_set_runstate((td), TD_EXITED); \
621 (td)->error = EFAULT; \
622 *__foo = 0; \
623 } \
624} while (0)
625
626static inline int fio_fill_issue_time(struct thread_data *td)
627{
628 if (td->o.read_iolog_file ||
629 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
630 return 1;
631
632 return 0;
633}
634
635static inline int __should_check_rate(struct thread_data *td,
636 enum fio_ddir ddir)
637{
638 struct thread_options *o = &td->o;
639
640 /*
641 * If some rate setting was given, we need to check it
642 */
643 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
644 o->rate_iops_min[ddir])
645 return 1;
646
647 return 0;
648}
649
650static inline int should_check_rate(struct thread_data *td)
651{
652 int ret = 0;
653
654 if (td->bytes_done[DDIR_READ])
655 ret |= __should_check_rate(td, DDIR_READ);
656 if (td->bytes_done[DDIR_WRITE])
657 ret |= __should_check_rate(td, DDIR_WRITE);
658 if (td->bytes_done[DDIR_TRIM])
659 ret |= __should_check_rate(td, DDIR_TRIM);
660
661 return ret;
662}
663
664static inline unsigned int td_max_bs(struct thread_data *td)
665{
666 unsigned int max_bs;
667
668 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
669 return max(td->o.max_bs[DDIR_TRIM], max_bs);
670}
671
672static inline unsigned int td_min_bs(struct thread_data *td)
673{
674 unsigned int min_bs;
675
676 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
677 return min(td->o.min_bs[DDIR_TRIM], min_bs);
678}
679
680static inline int td_async_processing(struct thread_data *td)
681{
682 return (td->flags & TD_F_NEED_LOCK) != 0;
683}
684
685/*
686 * We currently only need to do locking if we have verifier threads
687 * accessing our internal structures too
688 */
689static inline void td_io_u_lock(struct thread_data *td)
690{
691 if (td_async_processing(td))
692 pthread_mutex_lock(&td->io_u_lock);
693}
694
695static inline void td_io_u_unlock(struct thread_data *td)
696{
697 if (td_async_processing(td))
698 pthread_mutex_unlock(&td->io_u_lock);
699}
700
701static inline void td_io_u_free_notify(struct thread_data *td)
702{
703 if (td_async_processing(td))
704 pthread_cond_signal(&td->free_cond);
705}
706
707extern const char *fio_get_arch_string(int);
708extern const char *fio_get_os_string(int);
709
710#ifdef FIO_INTERNAL
711#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
712#define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
713#endif
714
715enum {
716 __FIO_OUTPUT_TERSE = 0,
717 __FIO_OUTPUT_JSON = 1,
718 __FIO_OUTPUT_NORMAL = 2,
719 __FIO_OUTPUT_JSON_PLUS = 3,
720 FIO_OUTPUT_NR = 4,
721
722 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
723 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
724 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
725 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
726};
727
728enum {
729 FIO_RAND_DIST_RANDOM = 0,
730 FIO_RAND_DIST_ZIPF,
731 FIO_RAND_DIST_PARETO,
732 FIO_RAND_DIST_GAUSS,
733 FIO_RAND_DIST_ZONED,
734};
735
736#define FIO_DEF_ZIPF 1.1
737#define FIO_DEF_PARETO 0.2
738
739enum {
740 FIO_RAND_GEN_TAUSWORTHE = 0,
741 FIO_RAND_GEN_LFSR,
742 FIO_RAND_GEN_TAUSWORTHE64,
743};
744
745enum {
746 FIO_CPUS_SHARED = 0,
747 FIO_CPUS_SPLIT,
748};
749
750extern void exec_trigger(const char *);
751extern void check_trigger_file(void);
752
753#endif
fio - Flexible IO Tester