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