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