projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Add support for trim as a workload 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
17struct thread_data;
18
19#include "compiler/compiler.h"
20#include "flist.h"
21#include "fifo.h"
22#include "rbtree.h"
23#include "arch/arch.h"
24#include "os/os.h"
25#include "mutex.h"
26#include "log.h"
27#include "debug.h"
28#include "file.h"
29#include "io_ddir.h"
30#include "ioengine.h"
31#include "iolog.h"
32#include "helpers.h"
33#include "options.h"
34#include "profile.h"
35#include "time.h"
36#include "gettime.h"
37#include "lib/getopt.h"
38#include "lib/rand.h"
39#include "server.h"
40#include "stat.h"
41#include "flow.h"
42
43#ifdef FIO_HAVE_GUASI
44#include <guasi.h>
45#endif
46
47#ifdef FIO_HAVE_SOLARISAIO
48#include <sys/asynch.h>
49#endif
50
51/*
52 * What type of allocation to use for io buffers
53 */
54enum fio_memtype {
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
60};
61
62/*
63 * offset generator types
64 */
65enum {
66 RW_SEQ_SEQ = 0,
67 RW_SEQ_IDENT,
68};
69
70/*
71 * What type of errors to continue on when continue_on_error is used
72 */
73enum error_type {
74 ERROR_TYPE_NONE = 0,
75 ERROR_TYPE_READ = 1 << 0,
76 ERROR_TYPE_WRITE = 1 << 1,
77 ERROR_TYPE_VERIFY = 1 << 2,
78 ERROR_TYPE_ANY = 0xffff,
79};
80
81struct bssplit {
82 unsigned int bs;
83 unsigned char perc;
84};
85
86struct thread_options {
87 int pad;
88 char *description;
89 char *name;
90 char *directory;
91 char *filename;
92 char *opendir;
93 char *ioengine;
94 enum td_ddir td_ddir;
95 unsigned int rw_seq;
96 unsigned int kb_base;
97 unsigned int ddir_seq_nr;
98 long ddir_seq_add;
99 unsigned int iodepth;
100 unsigned int iodepth_low;
101 unsigned int iodepth_batch;
102 unsigned int iodepth_batch_complete;
103
104 unsigned long long size;
105 unsigned int size_percent;
106 unsigned int fill_device;
107 unsigned long long file_size_low;
108 unsigned long long file_size_high;
109 unsigned long long start_offset;
110
111 unsigned int bs[DDIR_RWDIR_CNT];
112 unsigned int ba[DDIR_RWDIR_CNT];
113 unsigned int min_bs[DDIR_RWDIR_CNT];
114 unsigned int max_bs[DDIR_RWDIR_CNT];
115 struct bssplit *bssplit[DDIR_RWDIR_CNT];
116 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
117
118 unsigned int nr_files;
119 unsigned int open_files;
120 enum file_lock_mode file_lock_mode;
121 unsigned int lockfile_batch;
122
123 unsigned int odirect;
124 unsigned int invalidate_cache;
125 unsigned int create_serialize;
126 unsigned int create_fsync;
127 unsigned int create_on_open;
128 unsigned int create_only;
129 unsigned int end_fsync;
130 unsigned int pre_read;
131 unsigned int sync_io;
132 unsigned int verify;
133 unsigned int do_verify;
134 unsigned int verifysort;
135 unsigned int verify_interval;
136 unsigned int verify_offset;
137 char verify_pattern[MAX_PATTERN_SIZE];
138 unsigned int verify_pattern_bytes;
139 unsigned int verify_fatal;
140 unsigned int verify_dump;
141 unsigned int verify_async;
142 unsigned long long verify_backlog;
143 unsigned int verify_batch;
144 unsigned int use_thread;
145 unsigned int unlink;
146 unsigned int do_disk_util;
147 unsigned int override_sync;
148 unsigned int rand_repeatable;
149 unsigned int use_os_rand;
150 unsigned int write_lat_log;
151 unsigned int write_bw_log;
152 unsigned int write_iops_log;
153 unsigned int log_avg_msec;
154 unsigned int norandommap;
155 unsigned int softrandommap;
156 unsigned int bs_unaligned;
157 unsigned int fsync_on_close;
158
159 unsigned int hugepage_size;
160 unsigned int rw_min_bs;
161 unsigned int thinktime;
162 unsigned int thinktime_spin;
163 unsigned int thinktime_blocks;
164 unsigned int fsync_blocks;
165 unsigned int fdatasync_blocks;
166 unsigned int barrier_blocks;
167 unsigned long long start_delay;
168 unsigned long long timeout;
169 unsigned long long ramp_time;
170 unsigned int overwrite;
171 unsigned int bw_avg_time;
172 unsigned int iops_avg_time;
173 unsigned int loops;
174 unsigned long long zone_range;
175 unsigned long long zone_size;
176 unsigned long long zone_skip;
177 enum fio_memtype mem_type;
178 unsigned int mem_align;
179
180 unsigned int stonewall;
181 unsigned int new_group;
182 unsigned int numjobs;
183 os_cpu_mask_t cpumask;
184 unsigned int cpumask_set;
185 os_cpu_mask_t verify_cpumask;
186 unsigned int verify_cpumask_set;
187 unsigned int iolog;
188 unsigned int rwmixcycle;
189 unsigned int rwmix[2];
190 unsigned int nice;
191 unsigned int file_service_type;
192 unsigned int group_reporting;
193 unsigned int fadvise_hint;
194 enum fio_fallocate_mode fallocate_mode;
195 unsigned int zero_buffers;
196 unsigned int refill_buffers;
197 unsigned int scramble_buffers;
198 unsigned int compress_percentage;
199 unsigned int compress_chunk;
200 unsigned int time_based;
201 unsigned int disable_lat;
202 unsigned int disable_clat;
203 unsigned int disable_slat;
204 unsigned int disable_bw;
205 unsigned int gtod_reduce;
206 unsigned int gtod_cpu;
207 unsigned int gtod_offload;
208 enum fio_cs clocksource;
209 unsigned int no_stall;
210 unsigned int trim_percentage;
211 unsigned int trim_batch;
212 unsigned int trim_zero;
213 unsigned long long trim_backlog;
214 unsigned int clat_percentiles;
215 unsigned int overwrite_plist;
216 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
217
218 char *read_iolog_file;
219 char *write_iolog_file;
220 char *bw_log_file;
221 char *lat_log_file;
222 char *iops_log_file;
223 char *replay_redirect;
224
225 /*
226 * Pre-run and post-run shell
227 */
228 char *exec_prerun;
229 char *exec_postrun;
230
231 unsigned int rate[DDIR_RWDIR_CNT];
232 unsigned int ratemin[DDIR_RWDIR_CNT];
233 unsigned int ratecycle;
234 unsigned int rate_iops[DDIR_RWDIR_CNT];
235 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
236
237 char *ioscheduler;
238
239 /*
240 * CPU "io" cycle burner
241 */
242 unsigned int cpuload;
243 unsigned int cpucycle;
244
245 /*
246 * I/O Error handling
247 */
248 enum error_type continue_on_error;
249
250 /*
251 * Benchmark profile type
252 */
253 char *profile;
254
255 /*
256 * blkio cgroup support
257 */
258 char *cgroup;
259 unsigned int cgroup_weight;
260 unsigned int cgroup_nodelete;
261
262 unsigned int uid;
263 unsigned int gid;
264
265 int flow_id;
266 int flow;
267 int flow_watermark;
268 unsigned int flow_sleep;
269
270 unsigned long long offset_increment;
271
272 unsigned int sync_file_range;
273};
274
275/*
276 * This describes a single thread/process executing a fio job.
277 */
278struct thread_data {
279 struct thread_options o;
280 void *eo;
281 char verror[FIO_VERROR_SIZE];
282 pthread_t thread;
283 int thread_number;
284 int groupid;
285 struct thread_stat ts;
286
287 struct io_log *slat_log;
288 struct io_log *clat_log;
289 struct io_log *lat_log;
290 struct io_log *bw_log;
291 struct io_log *iops_log;
292
293 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
294 struct timeval bw_sample_time;
295
296 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
297 struct timeval iops_sample_time;
298
299 struct rusage ru_start;
300 struct rusage ru_end;
301
302 struct fio_file **files;
303 unsigned int files_size;
304 unsigned int files_index;
305 unsigned int nr_open_files;
306 unsigned int nr_done_files;
307 unsigned int nr_normal_files;
308 union {
309 unsigned int next_file;
310 os_random_state_t next_file_state;
311 struct frand_state __next_file_state;
312 };
313 int error;
314 int sig;
315 int done;
316 pid_t pid;
317 char *orig_buffer;
318 size_t orig_buffer_size;
319 volatile int terminate;
320 volatile int runstate;
321 unsigned int ioprio;
322 unsigned int ioprio_set;
323 unsigned int last_was_sync;
324 enum fio_ddir last_ddir;
325
326 char *mmapfile;
327 int mmapfd;
328
329 void *iolog_buf;
330 FILE *iolog_f;
331
332 char *sysfs_root;
333
334 unsigned long rand_seeds[8];
335
336 union {
337 os_random_state_t bsrange_state;
338 struct frand_state __bsrange_state;
339 };
340 union {
341 os_random_state_t verify_state;
342 struct frand_state __verify_state;
343 };
344 union {
345 os_random_state_t trim_state;
346 struct frand_state __trim_state;
347 };
348
349 struct frand_state buf_state;
350
351 unsigned int verify_batch;
352 unsigned int trim_batch;
353
354 int shm_id;
355
356 /*
357 * IO engine hooks, contains everything needed to submit an io_u
358 * to any of the available IO engines.
359 */
360 struct ioengine_ops *io_ops;
361
362 /*
363 * Queue depth of io_u's that fio MIGHT do
364 */
365 unsigned int cur_depth;
366
367 /*
368 * io_u's about to be committed
369 */
370 unsigned int io_u_queued;
371
372 /*
373 * io_u's submitted but not completed yet
374 */
375 unsigned int io_u_in_flight;
376
377 /*
378 * List of free and busy io_u's
379 */
380 struct flist_head io_u_freelist;
381 struct flist_head io_u_busylist;
382 struct flist_head io_u_requeues;
383 pthread_mutex_t io_u_lock;
384 pthread_cond_t free_cond;
385
386 /*
387 * async verify offload
388 */
389 struct flist_head verify_list;
390 pthread_t *verify_threads;
391 unsigned int nr_verify_threads;
392 pthread_cond_t verify_cond;
393 int verify_thread_exit;
394
395 /*
396 * Rate state
397 */
398 unsigned long long rate_bps[DDIR_RWDIR_CNT];
399 long rate_pending_usleep[DDIR_RWDIR_CNT];
400 unsigned long rate_bytes[DDIR_RWDIR_CNT];
401 unsigned long rate_blocks[DDIR_RWDIR_CNT];
402 struct timeval lastrate[DDIR_RWDIR_CNT];
403
404 unsigned long long total_io_size;
405 unsigned long long fill_device_size;
406
407 unsigned long io_issues[DDIR_RWDIR_CNT];
408 unsigned long long io_blocks[DDIR_RWDIR_CNT];
409 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
410 unsigned long long io_bytes[DDIR_RWDIR_CNT];
411 unsigned long long io_skip_bytes;
412 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
413 unsigned long long zone_bytes;
414 struct fio_mutex *mutex;
415
416 /*
417 * State for random io, a bitmap of blocks done vs not done
418 */
419 union {
420 os_random_state_t random_state;
421 struct frand_state __random_state;
422 };
423
424 struct timeval start; /* start of this loop */
425 struct timeval epoch; /* time job was started */
426 struct timeval last_issue;
427 struct timeval tv_cache;
428 unsigned int tv_cache_nr;
429 unsigned int tv_cache_mask;
430 unsigned int ramp_time_over;
431
432 /*
433 * read/write mixed workload state
434 */
435 union {
436 os_random_state_t rwmix_state;
437 struct frand_state __rwmix_state;
438 };
439 unsigned long rwmix_issues;
440 enum fio_ddir rwmix_ddir;
441 unsigned int ddir_seq_nr;
442
443 /*
444 * IO history logs for verification. We use a tree for sorting,
445 * if we are overwriting. Otherwise just use a fifo.
446 */
447 struct rb_root io_hist_tree;
448 struct flist_head io_hist_list;
449 unsigned long io_hist_len;
450
451 /*
452 * For IO replaying
453 */
454 struct flist_head io_log_list;
455
456 /*
457 * For tracking/handling discards
458 */
459 struct flist_head trim_list;
460 unsigned long trim_entries;
461
462 /*
463 * for fileservice, how often to switch to a new file
464 */
465 unsigned int file_service_nr;
466 unsigned int file_service_left;
467 struct fio_file *file_service_file;
468
469 unsigned int sync_file_range_nr;
470
471 /*
472 * For generating file sizes
473 */
474 union {
475 os_random_state_t file_size_state;
476 struct frand_state __file_size_state;
477 };
478
479 /*
480 * Error counts
481 */
482 unsigned int total_err_count;
483 int first_error;
484
485 struct fio_flow *flow;
486
487 /*
488 * Can be overloaded by profiles
489 */
490 struct prof_io_ops prof_io_ops;
491 void *prof_data;
492};
493
494/*
495 * when should interactive ETA output be generated
496 */
497enum {
498 FIO_ETA_AUTO,
499 FIO_ETA_ALWAYS,
500 FIO_ETA_NEVER,
501};
502
503#define __td_verror(td, err, msg, func) \
504 do { \
505 if ((td)->error) \
506 break; \
507 int e = (err); \
508 (td)->error = e; \
509 if (!(td)->first_error) \
510 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
511 } while (0)
512
513
514#define td_clear_error(td) \
515 (td)->error = 0;
516#define td_verror(td, err, func) \
517 __td_verror((td), (err), strerror((err)), (func))
518#define td_vmsg(td, err, msg, func) \
519 __td_verror((td), (err), (msg), (func))
520
521#define __fio_stringify_1(x) #x
522#define __fio_stringify(x) __fio_stringify_1(x)
523
524extern int exitall_on_terminate;
525extern unsigned int thread_number;
526extern unsigned int nr_process, nr_thread;
527extern int shm_id;
528extern int groupid;
529extern int terse_output;
530extern int temp_stall_ts;
531extern unsigned long long mlock_size;
532extern uintptr_t page_mask, page_size;
533extern int read_only;
534extern int eta_print;
535extern unsigned long done_secs;
536extern char *job_section;
537extern int fio_gtod_offload;
538extern int fio_gtod_cpu;
539extern enum fio_cs fio_clock_source;
540extern int warnings_fatal;
541extern int terse_version;
542extern int is_backend;
543extern int nr_clients;
544extern int log_syslog;
545extern const char fio_version_string[];
546extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
547
548extern struct thread_data *threads;
549
550static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
551{
552 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
553}
554
555#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
556#define TO_MAP_BLOCK(f, b) (b)
557#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
558#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
559
560#define REAL_MAX_JOBS 2048
561
562#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
563
564static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
565{
566 if (err == EILSEQ)
567 return ERROR_TYPE_VERIFY;
568 if (ddir == DDIR_READ)
569 return ERROR_TYPE_READ;
570 return ERROR_TYPE_WRITE;
571}
572
573static inline void update_error_count(struct thread_data *td, int err)
574{
575 td->total_err_count++;
576 if (td->total_err_count == 1)
577 td->first_error = err;
578}
579
580static inline int should_fsync(struct thread_data *td)
581{
582 if (td->last_was_sync)
583 return 0;
584 if (td->o.odirect)
585 return 0;
586 if (td_write(td) || td_rw(td) || td->o.override_sync)
587 return 1;
588
589 return 0;
590}
591
592/*
593 * Init/option functions
594 */
595extern int __must_check parse_options(int, char **);
596extern int parse_jobs_ini(char *, int, int);
597extern int parse_cmd_line(int, char **);
598extern int fio_backend(void);
599extern void reset_fio_state(void);
600extern void clear_io_state(struct thread_data *);
601extern int fio_options_parse(struct thread_data *, char **, int);
602extern void fio_keywords_init(void);
603extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
604extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
605extern void fio_fill_default_options(struct thread_data *);
606extern int fio_show_option_help(const char *);
607extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
608extern void fio_options_dup_and_init(struct option *);
609extern void fio_options_mem_dupe(struct thread_data *);
610extern void options_mem_dupe(void *data, struct fio_option *options);
611extern void td_fill_rand_seeds(struct thread_data *);
612extern void add_job_opts(const char **);
613extern char *num2str(unsigned long, int, int, int);
614extern int ioengine_load(struct thread_data *);
615
616#define FIO_GETOPT_JOB 0x89000000
617#define FIO_GETOPT_IOENGINE 0x98000000
618#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
619
620/*
621 * ETA/status stuff
622 */
623extern void print_thread_status(void);
624extern void print_status_init(int);
625
626/*
627 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
628 * will never back again. It may cycle between running/verififying/fsyncing.
629 * Once the thread reaches TD_EXITED, it is just waiting for the core to
630 * reap it.
631 */
632enum {
633 TD_NOT_CREATED = 0,
634 TD_CREATED,
635 TD_INITIALIZED,
636 TD_RAMP,
637 TD_RUNNING,
638 TD_PRE_READING,
639 TD_VERIFYING,
640 TD_FSYNCING,
641 TD_EXITED,
642 TD_REAPED,
643};
644
645extern void td_set_runstate(struct thread_data *, int);
646#define TERMINATE_ALL (-1)
647extern void fio_terminate_threads(int);
648
649/*
650 * Memory helpers
651 */
652extern int __must_check fio_pin_memory(void);
653extern void fio_unpin_memory(void);
654extern int __must_check allocate_io_mem(struct thread_data *);
655extern void free_io_mem(struct thread_data *);
656
657/*
658 * Reset stats after ramp time completes
659 */
660extern void reset_all_stats(struct thread_data *);
661
662/*
663 * blktrace support
664 */
665#ifdef FIO_HAVE_BLKTRACE
666extern int is_blktrace(const char *);
667extern int load_blktrace(struct thread_data *, const char *);
668#endif
669
670/*
671 * Mark unused variables passed to ops functions as unused, to silence gcc
672 */
673#define fio_unused __attribute((__unused__))
674#define fio_init __attribute__((constructor))
675#define fio_exit __attribute__((destructor))
676
677#define for_each_td(td, i) \
678 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
679#define for_each_file(td, f, i) \
680 if ((td)->files_index) \
681 for ((i) = 0, (f) = (td)->files[0]; \
682 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
683 (i)++)
684
685#define fio_assert(td, cond) do { \
686 if (!(cond)) { \
687 int *__foo = NULL; \
688 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
689 td_set_runstate((td), TD_EXITED); \
690 (td)->error = EFAULT; \
691 *__foo = 0; \
692 } \
693} while (0)
694
695static inline int fio_fill_issue_time(struct thread_data *td)
696{
697 if (td->o.read_iolog_file ||
698 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
699 return 1;
700
701 return 0;
702}
703
704static inline int __should_check_rate(struct thread_data *td,
705 enum fio_ddir ddir)
706{
707 struct thread_options *o = &td->o;
708
709 /*
710 * If some rate setting was given, we need to check it
711 */
712 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
713 o->rate_iops_min[ddir])
714 return 1;
715
716 return 0;
717}
718
719static inline int should_check_rate(struct thread_data *td,
720 unsigned long *bytes_done)
721{
722 int ret = 0;
723
724 if (bytes_done[DDIR_READ])
725 ret |= __should_check_rate(td, DDIR_READ);
726 if (bytes_done[DDIR_WRITE])
727 ret |= __should_check_rate(td, DDIR_WRITE);
728 if (bytes_done[DDIR_TRIM])
729 ret |= __should_check_rate(td, DDIR_TRIM);
730
731 return ret;
732}
733
734static inline int is_power_of_2(unsigned int val)
735{
736 return (val != 0 && ((val & (val - 1)) == 0));
737}
738
739/*
740 * We currently only need to do locking if we have verifier threads
741 * accessing our internal structures too
742 */
743static inline void td_io_u_lock(struct thread_data *td)
744{
745 if (td->o.verify_async)
746 pthread_mutex_lock(&td->io_u_lock);
747}
748
749static inline void td_io_u_unlock(struct thread_data *td)
750{
751 if (td->o.verify_async)
752 pthread_mutex_unlock(&td->io_u_lock);
753}
754
755static inline void td_io_u_free_notify(struct thread_data *td)
756{
757 if (td->o.verify_async)
758 pthread_cond_signal(&td->free_cond);
759}
760
761extern const char *fio_get_arch_string(int);
762extern const char *fio_get_os_string(int);
763
764#endif
fio - Flexible IO Tester