projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Start of support for auto-starting the backend
[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 "client.h"
40#include "server.h"
41#include "stat.h"
42#include "flow.h"
43
44#ifdef FIO_HAVE_GUASI
45#include <guasi.h>
46#endif
47
48#ifdef FIO_HAVE_SOLARISAIO
49#include <sys/asynch.h>
50#endif
51
52/*
53 * What type of allocation to use for io buffers
54 */
55enum fio_memtype {
56 MEM_MALLOC = 0, /* ordinary malloc */
57 MEM_SHM, /* use shared memory segments */
58 MEM_SHMHUGE, /* use shared memory segments with huge pages */
59 MEM_MMAP, /* use anonynomous mmap */
60 MEM_MMAPHUGE, /* memory mapped huge file */
61};
62
63/*
64 * offset generator types
65 */
66enum {
67 RW_SEQ_SEQ = 0,
68 RW_SEQ_IDENT,
69};
70
71/*
72 * What type of errors to continue on when continue_on_error is used
73 */
74enum error_type {
75 ERROR_TYPE_NONE = 0,
76 ERROR_TYPE_READ = 1 << 0,
77 ERROR_TYPE_WRITE = 1 << 1,
78 ERROR_TYPE_VERIFY = 1 << 2,
79 ERROR_TYPE_ANY = 0xffff,
80};
81
82struct bssplit {
83 unsigned int bs;
84 unsigned char perc;
85};
86
87struct thread_options {
88 int pad;
89 char *description;
90 char *name;
91 char *directory;
92 char *filename;
93 char *opendir;
94 char *ioengine;
95 enum td_ddir td_ddir;
96 unsigned int rw_seq;
97 unsigned int kb_base;
98 unsigned int ddir_seq_nr;
99 long ddir_seq_add;
100 unsigned int iodepth;
101 unsigned int iodepth_low;
102 unsigned int iodepth_batch;
103 unsigned int iodepth_batch_complete;
104
105 unsigned long long size;
106 unsigned int size_percent;
107 unsigned int fill_device;
108 unsigned long long file_size_low;
109 unsigned long long file_size_high;
110 unsigned long long start_offset;
111
112 unsigned int bs[2];
113 unsigned int ba[2];
114 unsigned int min_bs[2];
115 unsigned int max_bs[2];
116 struct bssplit *bssplit[2];
117 unsigned int bssplit_nr[2];
118
119 unsigned int nr_files;
120 unsigned int open_files;
121 enum file_lock_mode file_lock_mode;
122 unsigned int lockfile_batch;
123
124 unsigned int odirect;
125 unsigned int invalidate_cache;
126 unsigned int create_serialize;
127 unsigned int create_fsync;
128 unsigned int create_on_open;
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 time_based;
199 unsigned int disable_lat;
200 unsigned int disable_clat;
201 unsigned int disable_slat;
202 unsigned int disable_bw;
203 unsigned int gtod_reduce;
204 unsigned int gtod_cpu;
205 unsigned int gtod_offload;
206 enum fio_cs clocksource;
207 unsigned int no_stall;
208 unsigned int trim_percentage;
209 unsigned int trim_batch;
210 unsigned int trim_zero;
211 unsigned long long trim_backlog;
212 unsigned int clat_percentiles;
213 unsigned int overwrite_plist;
214 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
215
216 char *read_iolog_file;
217 char *write_iolog_file;
218 char *bw_log_file;
219 char *lat_log_file;
220 char *iops_log_file;
221 char *replay_redirect;
222
223 /*
224 * Pre-run and post-run shell
225 */
226 char *exec_prerun;
227 char *exec_postrun;
228
229 unsigned int rate[2];
230 unsigned int ratemin[2];
231 unsigned int ratecycle;
232 unsigned int rate_iops[2];
233 unsigned int rate_iops_min[2];
234
235 char *ioscheduler;
236
237 /*
238 * CPU "io" cycle burner
239 */
240 unsigned int cpuload;
241 unsigned int cpucycle;
242
243 /*
244 * I/O Error handling
245 */
246 enum error_type continue_on_error;
247
248 /*
249 * Benchmark profile type
250 */
251 char *profile;
252
253 /*
254 * blkio cgroup support
255 */
256 char *cgroup;
257 unsigned int cgroup_weight;
258 unsigned int cgroup_nodelete;
259
260 unsigned int uid;
261 unsigned int gid;
262
263 int flow_id;
264 int flow;
265 int flow_watermark;
266 unsigned int flow_sleep;
267
268 unsigned int sync_file_range;
269};
270
271/*
272 * This describes a single thread/process executing a fio job.
273 */
274struct thread_data {
275 struct thread_options o;
276 void *eo;
277 char verror[FIO_VERROR_SIZE];
278 pthread_t thread;
279 int thread_number;
280 int groupid;
281 struct thread_stat ts;
282
283 struct io_log *slat_log;
284 struct io_log *clat_log;
285 struct io_log *lat_log;
286 struct io_log *bw_log;
287 struct io_log *iops_log;
288
289 uint64_t stat_io_bytes[2];
290 struct timeval bw_sample_time;
291
292 uint64_t stat_io_blocks[2];
293 struct timeval iops_sample_time;
294
295 struct rusage ru_start;
296 struct rusage ru_end;
297
298 struct fio_file **files;
299 unsigned int files_size;
300 unsigned int files_index;
301 unsigned int nr_open_files;
302 unsigned int nr_done_files;
303 unsigned int nr_normal_files;
304 union {
305 unsigned int next_file;
306 os_random_state_t next_file_state;
307 struct frand_state __next_file_state;
308 };
309 int error;
310 int done;
311 pid_t pid;
312 char *orig_buffer;
313 size_t orig_buffer_size;
314 volatile int terminate;
315 volatile int runstate;
316 unsigned int ioprio;
317 unsigned int ioprio_set;
318 unsigned int last_was_sync;
319 enum fio_ddir last_ddir;
320
321 char *mmapfile;
322 int mmapfd;
323
324 void *iolog_buf;
325 FILE *iolog_f;
326
327 char *sysfs_root;
328
329 unsigned long rand_seeds[8];
330
331 union {
332 os_random_state_t bsrange_state;
333 struct frand_state __bsrange_state;
334 };
335 union {
336 os_random_state_t verify_state;
337 struct frand_state __verify_state;
338 };
339 union {
340 os_random_state_t trim_state;
341 struct frand_state __trim_state;
342 };
343
344 struct frand_state buf_state;
345
346 unsigned int verify_batch;
347 unsigned int trim_batch;
348
349 int shm_id;
350
351 /*
352 * IO engine hooks, contains everything needed to submit an io_u
353 * to any of the available IO engines.
354 */
355 struct ioengine_ops *io_ops;
356
357 /*
358 * Queue depth of io_u's that fio MIGHT do
359 */
360 unsigned int cur_depth;
361
362 /*
363 * io_u's about to be committed
364 */
365 unsigned int io_u_queued;
366
367 /*
368 * io_u's submitted but not completed yet
369 */
370 unsigned int io_u_in_flight;
371
372 /*
373 * List of free and busy io_u's
374 */
375 struct flist_head io_u_freelist;
376 struct flist_head io_u_busylist;
377 struct flist_head io_u_requeues;
378 pthread_mutex_t io_u_lock;
379 pthread_cond_t free_cond;
380
381 /*
382 * async verify offload
383 */
384 struct flist_head verify_list;
385 pthread_t *verify_threads;
386 unsigned int nr_verify_threads;
387 pthread_cond_t verify_cond;
388 int verify_thread_exit;
389
390 /*
391 * Rate state
392 */
393 unsigned long long rate_bps[2];
394 long rate_pending_usleep[2];
395 unsigned long rate_bytes[2];
396 unsigned long rate_blocks[2];
397 struct timeval lastrate[2];
398
399 unsigned long long total_io_size;
400 unsigned long long fill_device_size;
401
402 unsigned long io_issues[2];
403 unsigned long long io_blocks[2];
404 unsigned long long this_io_blocks[2];
405 unsigned long long io_bytes[2];
406 unsigned long long io_skip_bytes;
407 unsigned long long this_io_bytes[2];
408 unsigned long long zone_bytes;
409 struct fio_mutex *mutex;
410
411 /*
412 * State for random io, a bitmap of blocks done vs not done
413 */
414 union {
415 os_random_state_t random_state;
416 struct frand_state __random_state;
417 };
418
419 struct timeval start; /* start of this loop */
420 struct timeval epoch; /* time job was started */
421 struct timeval last_issue;
422 struct timeval tv_cache;
423 unsigned int tv_cache_nr;
424 unsigned int tv_cache_mask;
425 unsigned int ramp_time_over;
426
427 /*
428 * read/write mixed workload state
429 */
430 union {
431 os_random_state_t rwmix_state;
432 struct frand_state __rwmix_state;
433 };
434 unsigned long rwmix_issues;
435 enum fio_ddir rwmix_ddir;
436 unsigned int ddir_seq_nr;
437
438 /*
439 * IO history logs for verification. We use a tree for sorting,
440 * if we are overwriting. Otherwise just use a fifo.
441 */
442 struct rb_root io_hist_tree;
443 struct flist_head io_hist_list;
444 unsigned long io_hist_len;
445
446 /*
447 * For IO replaying
448 */
449 struct flist_head io_log_list;
450
451 /*
452 * For tracking/handling discards
453 */
454 struct flist_head trim_list;
455 unsigned long trim_entries;
456
457 /*
458 * for fileservice, how often to switch to a new file
459 */
460 unsigned int file_service_nr;
461 unsigned int file_service_left;
462 struct fio_file *file_service_file;
463
464 unsigned int sync_file_range_nr;
465
466 /*
467 * For generating file sizes
468 */
469 union {
470 os_random_state_t file_size_state;
471 struct frand_state __file_size_state;
472 };
473
474 /*
475 * Error counts
476 */
477 unsigned int total_err_count;
478 int first_error;
479
480 struct fio_flow *flow;
481
482 /*
483 * Can be overloaded by profiles
484 */
485 struct prof_io_ops prof_io_ops;
486 void *prof_data;
487};
488
489/*
490 * when should interactive ETA output be generated
491 */
492enum {
493 FIO_ETA_AUTO,
494 FIO_ETA_ALWAYS,
495 FIO_ETA_NEVER,
496};
497
498#define __td_verror(td, err, msg, func) \
499 do { \
500 if ((td)->error) \
501 break; \
502 int e = (err); \
503 (td)->error = e; \
504 if (!(td)->first_error) \
505 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
506 } while (0)
507
508
509#define td_clear_error(td) \
510 (td)->error = 0;
511#define td_verror(td, err, func) \
512 __td_verror((td), (err), strerror((err)), (func))
513#define td_vmsg(td, err, msg, func) \
514 __td_verror((td), (err), (msg), (func))
515
516#define __fio_stringify_1(x) #x
517#define __fio_stringify(x) __fio_stringify_1(x)
518
519extern int exitall_on_terminate;
520extern unsigned int thread_number;
521extern unsigned int nr_process, nr_thread;
522extern int shm_id;
523extern int groupid;
524extern int terse_output;
525extern int temp_stall_ts;
526extern unsigned long long mlock_size;
527extern unsigned long page_mask, page_size;
528extern int read_only;
529extern int eta_print;
530extern unsigned long done_secs;
531extern char *job_section;
532extern int fio_gtod_offload;
533extern int fio_gtod_cpu;
534extern enum fio_cs fio_clock_source;
535extern int warnings_fatal;
536extern int terse_version;
537extern int is_backend;
538extern int nr_clients;
539extern int log_syslog;
540extern const char fio_version_string[];
541extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
542
543extern struct thread_data *threads;
544
545static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
546{
547 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
548}
549
550#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
551#define TO_MAP_BLOCK(f, b) (b)
552#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
553#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
554
555#define REAL_MAX_JOBS 2048
556
557#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
558
559static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
560{
561 if (err == EILSEQ)
562 return ERROR_TYPE_VERIFY;
563 if (ddir == DDIR_READ)
564 return ERROR_TYPE_READ;
565 return ERROR_TYPE_WRITE;
566}
567
568static inline void update_error_count(struct thread_data *td, int err)
569{
570 td->total_err_count++;
571 if (td->total_err_count == 1)
572 td->first_error = err;
573}
574
575static inline int should_fsync(struct thread_data *td)
576{
577 if (td->last_was_sync)
578 return 0;
579 if (td->o.odirect)
580 return 0;
581 if (td_write(td) || td_rw(td) || td->o.override_sync)
582 return 1;
583
584 return 0;
585}
586
587/*
588 * Init/option functions
589 */
590extern int __must_check fio_init_options(void);
591extern int __must_check parse_options(int, char **);
592extern int parse_jobs_ini(char *, int, int);
593extern int parse_cmd_line(int, char **);
594extern int fio_backend(void);
595extern void reset_fio_state(void);
596extern void clear_io_state(struct thread_data *);
597extern int fio_options_parse(struct thread_data *, char **, int);
598extern void fio_keywords_init(void);
599extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
600extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
601extern void fio_fill_default_options(struct thread_data *);
602extern int fio_show_option_help(const char *);
603extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
604extern void fio_options_dup_and_init(struct option *);
605extern void fio_options_mem_dupe(struct thread_data *);
606extern void options_mem_dupe(void *data, struct fio_option *options);
607extern void td_fill_rand_seeds(struct thread_data *);
608extern void add_job_opts(const char **);
609extern char *num2str(unsigned long, int, int, int);
610extern int ioengine_load(struct thread_data *);
611
612extern unsigned long page_mask;
613extern unsigned long page_size;
614extern int initialize_fio(char *envp[]);
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[0])
725 ret |= __should_check_rate(td, 0);
726 if (bytes_done[1])
727 ret |= __should_check_rate(td, 1);
728
729 return ret;
730}
731
732static inline int is_power_of_2(unsigned int val)
733{
734 return (val != 0 && ((val & (val - 1)) == 0));
735}
736
737/*
738 * We currently only need to do locking if we have verifier threads
739 * accessing our internal structures too
740 */
741static inline void td_io_u_lock(struct thread_data *td)
742{
743 if (td->o.verify_async)
744 pthread_mutex_lock(&td->io_u_lock);
745}
746
747static inline void td_io_u_unlock(struct thread_data *td)
748{
749 if (td->o.verify_async)
750 pthread_mutex_unlock(&td->io_u_lock);
751}
752
753static inline void td_io_u_free_notify(struct thread_data *td)
754{
755 if (td->o.verify_async)
756 pthread_cond_signal(&td->free_cond);
757}
758
759extern const char *fio_get_arch_string(int);
760extern const char *fio_get_os_string(int);
761
762#endif
fio - Flexible IO Tester