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