projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Get rid of reopen_files()
[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
15#include "list.h"
16#include "md5.h"
17#include "crc32.h"
18#include "arch.h"
19#include "os.h"
20
21#ifdef FIO_HAVE_SYSLET
22#include "syslet.h"
23#endif
24
25enum fio_ddir {
26 DDIR_READ = 0,
27 DDIR_WRITE,
28 DDIR_SYNC,
29};
30
31enum td_ddir {
32 TD_DDIR_READ = 1 << 0,
33 TD_DDIR_WRITE = 1 << 1,
34 TD_DDIR_RAND = 1 << 2,
35 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
36 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
37 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
38 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
39};
40
41/*
42 * Use for maintaining statistics
43 */
44struct io_stat {
45 unsigned long max_val;
46 unsigned long min_val;
47 unsigned long samples;
48
49 double mean;
50 double S;
51};
52
53/*
54 * A single data sample
55 */
56struct io_sample {
57 unsigned long time;
58 unsigned long val;
59 enum fio_ddir ddir;
60};
61
62/*
63 * Dynamically growing data sample log
64 */
65struct io_log {
66 unsigned long nr_samples;
67 unsigned long max_samples;
68 struct io_sample *log;
69};
70
71/*
72 * When logging io actions, this matches a single sent io_u
73 */
74struct io_piece {
75 struct list_head list;
76 struct fio_file *file;
77 unsigned long long offset;
78 unsigned long len;
79 enum fio_ddir ddir;
80};
81
82#ifdef FIO_HAVE_SYSLET
83struct syslet_req {
84 struct syslet_uatom atom; /* the atom to submit */
85 struct syslet_uatom *head; /* head of the sequence */
86 long ret; /* syscall return value */
87};
88#endif
89
90enum {
91 IO_U_F_FREE = 1 << 0,
92 IO_U_F_FLIGHT = 1 << 1,
93};
94
95/*
96 * The io unit
97 */
98struct io_u {
99 union {
100#ifdef FIO_HAVE_LIBAIO
101 struct iocb iocb;
102#endif
103#ifdef FIO_HAVE_POSIXAIO
104 struct aiocb aiocb;
105#endif
106#ifdef FIO_HAVE_SGIO
107 struct sg_io_hdr hdr;
108#endif
109#ifdef FIO_HAVE_SYSLET
110 struct syslet_req req;
111#endif
112 };
113 struct timeval start_time;
114 struct timeval issue_time;
115
116 /*
117 * Allocated/set buffer and length
118 */
119 void *buf;
120 unsigned long buflen;
121 unsigned long long offset;
122
123 /*
124 * IO engine state, may be different from above when we get
125 * partial transfers / residual data counts
126 */
127 void *xfer_buf;
128 unsigned long xfer_buflen;
129
130 unsigned int resid;
131 unsigned int error;
132
133 enum fio_ddir ddir;
134
135 /*
136 * io engine private data
137 */
138 union {
139 unsigned int index;
140 unsigned int seen;
141 };
142
143 unsigned int flags;
144
145 struct fio_file *file;
146
147 struct list_head list;
148
149 /*
150 * Callback for io completion
151 */
152 int (*end_io)(struct io_u *);
153};
154
155/*
156 * io_ops->queue() return values
157 */
158enum {
159 FIO_Q_COMPLETED = 0, /* completed sync */
160 FIO_Q_QUEUED = 1, /* queued, will complete async */
161 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
162};
163
164#define FIO_HDR_MAGIC 0xf00baaef
165
166enum {
167 VERIFY_NONE = 0, /* no verification */
168 VERIFY_MD5, /* md5 sum data blocks */
169 VERIFY_CRC32, /* crc32 sum data blocks */
170};
171
172/*
173 * A header structure associated with each checksummed data block
174 */
175struct verify_header {
176 unsigned int fio_magic;
177 unsigned int len;
178 unsigned int verify_type;
179 union {
180 char md5_digest[MD5_HASH_WORDS * 4];
181 unsigned long crc32;
182 };
183};
184
185struct group_run_stats {
186 unsigned long long max_run[2], min_run[2];
187 unsigned long long max_bw[2], min_bw[2];
188 unsigned long long io_kb[2];
189 unsigned long long agg[2];
190};
191
192/*
193 * What type of allocation to use for io buffers
194 */
195enum fio_memtype {
196 MEM_MALLOC = 0, /* ordinary malloc */
197 MEM_SHM, /* use shared memory segments */
198 MEM_SHMHUGE, /* use shared memory segments with huge pages */
199 MEM_MMAP, /* use anonynomous mmap */
200 MEM_MMAPHUGE, /* memory mapped huge file */
201};
202
203/*
204 * The type of object we are working on
205 */
206enum fio_filetype {
207 FIO_TYPE_FILE = 1, /* plain file */
208 FIO_TYPE_BD, /* block device */
209 FIO_TYPE_CHAR, /* character device */
210};
211
212enum fio_ioengine_flags {
213 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
214 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
215 FIO_RAWIO = 1 << 2, /* some sort of direct/raw io */
216 FIO_DISKLESSIO = 1 << 3, /* no disk involved */
217};
218
219/*
220 * Each thread_data structure has a number of files associated with it,
221 * this structure holds state information for a single file.
222 */
223struct fio_file {
224 /*
225 * A file may not be a file descriptor, let the io engine decide
226 */
227 union {
228 unsigned long file_data;
229 int fd;
230 };
231 char *file_name;
232 void *mmap;
233 unsigned long long file_size;
234 unsigned long long real_file_size;
235 unsigned long long file_offset;
236 unsigned long long last_pos;
237 unsigned long long last_completed_pos;
238
239 /*
240 * block map for random io
241 */
242 unsigned long *file_map;
243 unsigned int num_maps;
244 unsigned int last_free_lookup;
245
246 unsigned int unlink;
247 unsigned int open;
248};
249
250/*
251 * How many depth levels to log
252 */
253#define FIO_IO_U_MAP_NR 8
254#define FIO_IO_U_LAT_NR 12
255
256struct thread_stat {
257 char *name;
258 char *verror;
259 int error;
260 int groupid;
261 pid_t pid;
262 char *description;
263 int members;
264
265 struct io_log *slat_log;
266 struct io_log *clat_log;
267 struct io_log *bw_log;
268
269 /*
270 * bandwidth and latency stats
271 */
272 struct io_stat clat_stat[2]; /* completion latency */
273 struct io_stat slat_stat[2]; /* submission latency */
274 struct io_stat bw_stat[2]; /* bandwidth stats */
275
276 unsigned long long stat_io_bytes[2];
277 struct timeval stat_sample_time[2];
278
279 /*
280 * fio system usage accounting
281 */
282 struct rusage ru_start;
283 struct rusage ru_end;
284 unsigned long usr_time;
285 unsigned long sys_time;
286 unsigned long ctx;
287
288 /*
289 * IO depth and latency stats
290 */
291 unsigned int io_u_map[FIO_IO_U_MAP_NR];
292 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
293 unsigned long total_io_u;
294
295 unsigned long long io_bytes[2];
296 unsigned long runtime[2];
297 unsigned long total_run_time;
298};
299
300/*
301 * This describes a single thread/process executing a fio job.
302 */
303struct thread_data {
304 int pad;
305 char *description;
306 char *name;
307 char *directory;
308 char *filename;
309 char verror[128];
310 pthread_t thread;
311 int thread_number;
312 int groupid;
313 struct thread_stat ts;
314 enum fio_filetype filetype;
315 struct fio_file *files;
316 unsigned int nr_files;
317 unsigned int nr_open_files;
318 unsigned int nr_uniq_files;
319 union {
320 unsigned int next_file;
321 os_random_state_t next_file_state;
322 };
323 int error;
324 pid_t pid;
325 char *orig_buffer;
326 size_t orig_buffer_size;
327 volatile int terminate;
328 volatile int runstate;
329 enum td_ddir td_ddir;
330 unsigned int ioprio;
331 unsigned int last_was_sync;
332
333 unsigned int odirect;
334 unsigned int invalidate_cache;
335 unsigned int create_serialize;
336 unsigned int create_fsync;
337 unsigned int end_fsync;
338 unsigned int sync_io;
339 unsigned int verify;
340 unsigned int use_thread;
341 unsigned int unlink;
342 unsigned int do_disk_util;
343 unsigned int override_sync;
344 unsigned int rand_repeatable;
345 unsigned int write_lat_log;
346 unsigned int write_bw_log;
347 unsigned int norandommap;
348 unsigned int bs_unaligned;
349
350 unsigned int bs[2];
351 unsigned int min_bs[2];
352 unsigned int max_bs[2];
353 unsigned int hugepage_size;
354 unsigned int rw_min_bs;
355 unsigned int thinktime;
356 unsigned int thinktime_spin;
357 unsigned int thinktime_blocks;
358 unsigned int fsync_blocks;
359 unsigned int start_delay;
360 unsigned long timeout;
361 unsigned int overwrite;
362 unsigned int bw_avg_time;
363 unsigned int loops;
364 unsigned long long zone_size;
365 unsigned long long zone_skip;
366 enum fio_memtype mem_type;
367 char *mmapfile;
368 int mmapfd;
369 unsigned int stonewall;
370 unsigned int numjobs;
371 unsigned int iodepth;
372 unsigned int iodepth_low;
373 unsigned int iodepth_batch;
374 os_cpu_mask_t cpumask;
375 unsigned int iolog;
376 unsigned int read_iolog;
377 unsigned int rwmixcycle;
378 unsigned int rwmixread;
379 unsigned int rwmixwrite;
380 unsigned int nice;
381 unsigned int file_service_type;
382 unsigned int group_reporting;
383 unsigned int open_files;
384
385 char *read_iolog_file;
386 char *write_iolog_file;
387 void *iolog_buf;
388 FILE *iolog_f;
389
390 char *sysfs_root;
391 char *ioscheduler;
392
393 os_random_state_t bsrange_state;
394 os_random_state_t verify_state;
395
396 int shm_id;
397
398 /*
399 * IO engine hooks, contains everything needed to submit an io_u
400 * to any of the available IO engines.
401 */
402 struct ioengine_ops *io_ops;
403
404 /*
405 * Current IO depth and list of free and busy io_u's.
406 */
407 unsigned int cur_depth;
408 struct list_head io_u_freelist;
409 struct list_head io_u_busylist;
410 struct list_head io_u_requeues;
411 unsigned int io_u_queued;
412
413 /*
414 * Rate state
415 */
416 unsigned int rate;
417 unsigned int ratemin;
418 unsigned int ratecycle;
419 unsigned long rate_usec_cycle;
420 long rate_pending_usleep;
421 unsigned long rate_bytes;
422 struct timeval lastrate;
423
424 unsigned long long io_size;
425 unsigned long long total_file_size;
426 unsigned long long start_offset;
427 unsigned long long total_io_size;
428
429 unsigned long io_issues[2];
430 unsigned long long io_blocks[2];
431 unsigned long long io_bytes[2];
432 unsigned long long this_io_bytes[2];
433 unsigned long long zone_bytes;
434 volatile int mutex;
435
436 /*
437 * State for random io, a bitmap of blocks done vs not done
438 */
439 os_random_state_t random_state;
440
441 /*
442 * CPU "io" cycle burner
443 */
444 unsigned int cpuload;
445 unsigned int cpucycle;
446
447 struct timeval start; /* start of this loop */
448 struct timeval epoch; /* time job was started */
449
450 /*
451 * read/write mixed workload state
452 */
453 os_random_state_t rwmix_state;
454 struct timeval rwmix_switch;
455 enum fio_ddir rwmix_ddir;
456
457 /*
458 * Pre-run and post-run shell
459 */
460 char *exec_prerun;
461 char *exec_postrun;
462
463 /*
464 * IO historic logs
465 */
466 struct list_head io_hist_list;
467 struct list_head io_log_list;
468
469 /*
470 * timeout handling
471 */
472 struct timeval timeout_end;
473 struct itimerval timer;
474};
475
476/*
477 * roundrobin available files, or choose one at random.
478 */
479enum {
480 FIO_FSERVICE_RANDOM = 1,
481 FIO_FSERVICE_RR = 2,
482};
483
484/*
485 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
486 * the timer on each queue operation.
487 */
488#define IO_U_TIMEOUT_INC 5
489#define IO_U_TIMEOUT 30
490
491#define __td_verror(td, err, msg, func) \
492 do { \
493 if ((td)->error) \
494 break; \
495 int e = (err); \
496 (td)->error = e; \
497 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
498 } while (0)
499
500
501#define td_verror(td, err, func) \
502 __td_verror((td), (err), strerror((err)), (func))
503#define td_vmsg(td, err, msg, func) \
504 __td_verror((td), (err), (msg), (func))
505
506extern int exitall_on_terminate;
507extern int thread_number;
508extern int shm_id;
509extern int groupid;
510extern int terse_output;
511extern FILE *f_out;
512extern FILE *f_err;
513extern int temp_stall_ts;
514extern unsigned long long mlock_size;
515
516extern struct thread_data *threads;
517
518#define td_read(td) ((td)->td_ddir & TD_DDIR_READ)
519#define td_write(td) ((td)->td_ddir & TD_DDIR_WRITE)
520#define td_rw(td) (((td)->td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
521#define td_random(td) ((td)->td_ddir & TD_DDIR_RAND)
522
523#define BLOCKS_PER_MAP (8 * sizeof(long))
524#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
525#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
526#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
527
528#define MAX_JOBS (1024)
529
530static inline int should_fsync(struct thread_data *td)
531{
532 if (td->last_was_sync)
533 return 0;
534 if (td->odirect)
535 return 0;
536 if (td_write(td) || td_rw(td) || td->override_sync)
537 return 1;
538
539 return 0;
540}
541
542/*
543 * Disk utils as read in /sys/block/<dev>/stat
544 */
545struct disk_util_stat {
546 unsigned ios[2];
547 unsigned merges[2];
548 unsigned long long sectors[2];
549 unsigned ticks[2];
550 unsigned io_ticks;
551 unsigned time_in_queue;
552};
553
554/*
555 * Per-device disk util management
556 */
557struct disk_util {
558 struct list_head list;
559
560 char *name;
561 char path[256];
562 dev_t dev;
563
564 struct disk_util_stat dus;
565 struct disk_util_stat last_dus;
566
567 unsigned long msec;
568 struct timeval time;
569};
570
571#define DISK_UTIL_MSEC (250)
572
573#ifndef min
574#define min(a, b) ((a) < (b) ? (a) : (b))
575#endif
576#ifndef max
577#define max(a, b) ((a) > (b) ? (a) : (b))
578#endif
579
580/*
581 * Log exports
582 */
583extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
584extern void write_iolog_put(struct thread_data *, struct io_u *);
585extern int __must_check init_iolog(struct thread_data *td);
586extern void log_io_piece(struct thread_data *, struct io_u *);
587extern void prune_io_piece_log(struct thread_data *);
588extern void write_iolog_close(struct thread_data *);
589
590/*
591 * Logging
592 */
593extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
594extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
595extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
596extern void show_run_stats(void);
597extern void init_disk_util(struct thread_data *);
598extern void update_rusage_stat(struct thread_data *);
599extern void update_io_ticks(void);
600extern void disk_util_timer_arm(void);
601extern void setup_log(struct io_log **);
602extern void finish_log(struct thread_data *, struct io_log *, const char *);
603extern void __finish_log(struct io_log *, const char *);
604extern int setup_rate(struct thread_data *);
605extern struct io_log *agg_io_log[2];
606extern int write_bw_log;
607extern void add_agg_sample(unsigned long, enum fio_ddir);
608
609/*
610 * Time functions
611 */
612extern unsigned long utime_since(struct timeval *, struct timeval *);
613extern unsigned long utime_since_now(struct timeval *);
614extern unsigned long mtime_since(struct timeval *, struct timeval *);
615extern unsigned long mtime_since_now(struct timeval *);
616extern unsigned long time_since_now(struct timeval *);
617extern unsigned long mtime_since_genesis(void);
618extern void __usec_sleep(unsigned int);
619extern void usec_sleep(struct thread_data *, unsigned long);
620extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
621extern void fill_start_time(struct timeval *);
622extern void fio_gettime(struct timeval *, void *);
623extern void set_genesis_time(void);
624
625/*
626 * Init functions
627 */
628extern int __must_check parse_options(int, char **);
629extern int __must_check init_random_state(struct thread_data *);
630
631/*
632 * File setup/shutdown
633 */
634extern void close_files(struct thread_data *);
635extern int __must_check setup_files(struct thread_data *);
636extern int __must_check open_files(struct thread_data *);
637extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
638extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
639extern void generic_close_file(struct thread_data *, struct fio_file *);
640
641/*
642 * ETA/status stuff
643 */
644extern void print_thread_status(void);
645extern void print_status_init(int);
646
647/*
648 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
649 * will never back again. It may cycle between running/verififying/fsyncing.
650 * Once the thread reaches TD_EXITED, it is just waiting for the core to
651 * reap it.
652 */
653enum {
654 TD_NOT_CREATED = 0,
655 TD_CREATED,
656 TD_INITIALIZED,
657 TD_RUNNING,
658 TD_VERIFYING,
659 TD_FSYNCING,
660 TD_EXITED,
661 TD_REAPED,
662};
663
664/*
665 * Verify helpers
666 */
667extern void populate_verify_io_u(struct thread_data *, struct io_u *);
668extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
669extern int __must_check verify_io_u(struct io_u *);
670
671/*
672 * Memory helpers
673 */
674extern int __must_check fio_pin_memory(void);
675extern void fio_unpin_memory(void);
676extern int __must_check allocate_io_mem(struct thread_data *);
677extern void free_io_mem(struct thread_data *);
678
679/*
680 * io unit handling
681 */
682#define queue_full(td) list_empty(&(td)->io_u_freelist)
683extern struct io_u *__get_io_u(struct thread_data *);
684extern struct io_u *get_io_u(struct thread_data *);
685extern void put_io_u(struct thread_data *, struct io_u *);
686extern void requeue_io_u(struct thread_data *, struct io_u **);
687extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
688extern long __must_check io_u_queued_complete(struct thread_data *, int);
689extern void io_u_queued(struct thread_data *, struct io_u *);
690extern void io_u_log_error(struct thread_data *, struct io_u *);
691extern void io_u_init_timeout(void);
692extern void io_u_set_timeout(struct thread_data *);
693
694/*
695 * io engine entry points
696 */
697extern int __must_check td_io_init(struct thread_data *);
698extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
699extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
700extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
701extern int __must_check td_io_getevents(struct thread_data *, int, int, struct timespec *);
702extern int __must_check td_io_commit(struct thread_data *);
703extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
704extern void td_io_close_file(struct thread_data *, struct fio_file *);
705
706/*
707 * This is a pretty crappy semaphore implementation, but with the use that fio
708 * has (just signalling start/go conditions), it doesn't have to be better.
709 * Naturally this would not work for any type of contended semaphore or
710 * for real locking.
711 */
712static inline void fio_sem_init(volatile int *sem, int val)
713{
714 *sem = val;
715}
716
717static inline void fio_sem_down(volatile int *sem)
718{
719 while (*sem == 0)
720 usleep(10000);
721
722 (*sem)--;
723}
724
725static inline void fio_sem_up(volatile int *sem)
726{
727 (*sem)++;
728}
729
730/*
731 * If logging output to a file, stderr should go to both stderr and f_err
732 */
733#define log_err(args...) do { \
734 fprintf(f_err, ##args); \
735 if (f_err != stderr) \
736 fprintf(stderr, ##args); \
737 } while (0)
738
739FILE *get_f_out(void);
740FILE *get_f_err(void);
741
742struct ioengine_ops {
743 struct list_head list;
744 char name[16];
745 int version;
746 int flags;
747 int (*setup)(struct thread_data *);
748 int (*init)(struct thread_data *);
749 int (*prep)(struct thread_data *, struct io_u *);
750 int (*queue)(struct thread_data *, struct io_u *);
751 int (*commit)(struct thread_data *);
752 int (*getevents)(struct thread_data *, int, int, struct timespec *);
753 struct io_u *(*event)(struct thread_data *, int);
754 int (*cancel)(struct thread_data *, struct io_u *);
755 void (*cleanup)(struct thread_data *);
756 int (*open_file)(struct thread_data *, struct fio_file *);
757 void (*close_file)(struct thread_data *, struct fio_file *);
758 void *data;
759 void *dlhandle;
760};
761
762#define FIO_IOOPS_VERSION 6
763
764extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
765extern void register_ioengine(struct ioengine_ops *);
766extern void unregister_ioengine(struct ioengine_ops *);
767extern void close_ioengine(struct thread_data *);
768
769/*
770 * Mark unused variables passed to ops functions as unused, to silence gcc
771 */
772#define fio_unused __attribute((__unused__))
773#define fio_init __attribute__((constructor))
774#define fio_exit __attribute__((destructor))
775
776#define for_each_td(td, i) \
777 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
778#define for_each_file(td, f, i) \
779 for ((i) = 0, (f) = &(td)->files[0]; (i) < (int) (td)->open_files; (i)++, (f)++)
780
781#define fio_assert(td, cond) do { \
782 if (!(cond)) { \
783 int *__foo = NULL; \
784 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
785 (td)->runstate = TD_EXITED; \
786 (td)->error = EFAULT; \
787 *__foo = 0; \
788 } \
789} while (0)
790
791#endif
fio - Flexible IO Tester