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