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