Fix file open leak
[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 <getopt.h>
15#include <inttypes.h>
16#include <assert.h>
17
18#include "compiler/compiler.h"
19#include "flist.h"
20#include "fifo.h"
21#include "rbtree.h"
22#include "arch/arch.h"
23#include "os/os.h"
24#include "mutex.h"
25#include "log.h"
26#include "debug.h"
27
28#ifdef FIO_HAVE_GUASI
29#include <guasi.h>
30#endif
31
32#ifdef FIO_HAVE_SOLARISAIO
33#include <sys/asynch.h>
34#endif
35
36enum fio_ddir {
37 DDIR_READ = 0,
38 DDIR_WRITE,
39 DDIR_SYNC,
40 DDIR_INVAL = -1,
41};
42
43enum td_ddir {
44 TD_DDIR_READ = 1 << 0,
45 TD_DDIR_WRITE = 1 << 1,
46 TD_DDIR_RAND = 1 << 2,
47 TD_DDIR_RW = TD_DDIR_READ | TD_DDIR_WRITE,
48 TD_DDIR_RANDREAD = TD_DDIR_READ | TD_DDIR_RAND,
49 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
50 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
51};
52
53enum file_lock_mode {
54 FILE_LOCK_NONE,
55 FILE_LOCK_EXCLUSIVE,
56 FILE_LOCK_READWRITE,
57};
58
59/*
60 * Use for maintaining statistics
61 */
62struct io_stat {
63 unsigned long max_val;
64 unsigned long min_val;
65 unsigned long samples;
66
67 double mean;
68 double S;
69};
70
71/*
72 * A single data sample
73 */
74struct io_sample {
75 unsigned long time;
76 unsigned long val;
77 enum fio_ddir ddir;
78};
79
80/*
81 * Dynamically growing data sample log
82 */
83struct io_log {
84 unsigned long nr_samples;
85 unsigned long max_samples;
86 struct io_sample *log;
87};
88
89/*
90 * When logging io actions, this matches a single sent io_u
91 */
92struct io_piece {
93 union {
94 struct rb_node rb_node;
95 struct flist_head list;
96 };
97 union {
98 int fileno;
99 struct fio_file *file;
100 };
101 unsigned long long offset;
102 unsigned long len;
103 enum fio_ddir ddir;
104 union {
105 unsigned long delay;
106 unsigned int file_action;
107 };
108};
109
110enum {
111 IO_U_F_FREE = 1 << 0,
112 IO_U_F_FLIGHT = 1 << 1,
113};
114
115struct thread_data;
116
117/*
118 * The io unit
119 */
120struct io_u {
121 union {
122#ifdef FIO_HAVE_LIBAIO
123 struct iocb iocb;
124#endif
125#ifdef FIO_HAVE_POSIXAIO
126 struct aiocb aiocb;
127#endif
128#ifdef FIO_HAVE_SGIO
129 struct sg_io_hdr hdr;
130#endif
131#ifdef FIO_HAVE_GUASI
132 guasi_req_t greq;
133#endif
134#ifdef FIO_HAVE_SOLARISAIO
135 aio_result_t resultp;
136#endif
137 };
138 struct timeval start_time;
139 struct timeval issue_time;
140
141 /*
142 * Allocated/set buffer and length
143 */
144 void *buf;
145 unsigned long buflen;
146 unsigned long long offset;
147 unsigned long long endpos;
148
149 /*
150 * IO engine state, may be different from above when we get
151 * partial transfers / residual data counts
152 */
153 void *xfer_buf;
154 unsigned long xfer_buflen;
155
156 unsigned int resid;
157 unsigned int error;
158
159 enum fio_ddir ddir;
160
161 /*
162 * io engine private data
163 */
164 union {
165 unsigned int index;
166 unsigned int seen;
167 void *engine_data;
168 };
169
170 unsigned int flags;
171
172 struct fio_file *file;
173
174 struct flist_head list;
175
176 /*
177 * Callback for io completion
178 */
179 int (*end_io)(struct thread_data *, struct io_u *);
180};
181
182/*
183 * io_ops->queue() return values
184 */
185enum {
186 FIO_Q_COMPLETED = 0, /* completed sync */
187 FIO_Q_QUEUED = 1, /* queued, will complete async */
188 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
189};
190
191#define FIO_HDR_MAGIC 0xf00baaef
192
193enum {
194 VERIFY_NONE = 0, /* no verification */
195 VERIFY_MD5, /* md5 sum data blocks */
196 VERIFY_CRC64, /* crc64 sum data blocks */
197 VERIFY_CRC32, /* crc32 sum data blocks */
198 VERIFY_CRC32C, /* crc32c sum data blocks */
199 VERIFY_CRC32C_INTEL, /* crc32c sum data blocks with hw */
200 VERIFY_CRC16, /* crc16 sum data blocks */
201 VERIFY_CRC7, /* crc7 sum data blocks */
202 VERIFY_SHA256, /* sha256 sum data blocks */
203 VERIFY_SHA512, /* sha512 sum data blocks */
204 VERIFY_META, /* block_num, timestamp etc. */
205 VERIFY_NULL, /* pretend to verify */
206};
207
208/*
209 * A header structure associated with each checksummed data block. It is
210 * followed by a checksum specific header that contains the verification
211 * data.
212 */
213struct verify_header {
214 unsigned int fio_magic;
215 unsigned int len;
216 unsigned int verify_type;
217};
218
219struct vhdr_md5 {
220 uint32_t md5_digest[16];
221};
222struct vhdr_sha512 {
223 uint8_t sha512[128];
224};
225struct vhdr_sha256 {
226 uint8_t sha256[128];
227};
228struct vhdr_crc64 {
229 uint64_t crc64;
230};
231struct vhdr_crc32 {
232 uint32_t crc32;
233};
234struct vhdr_crc16 {
235 uint16_t crc16;
236};
237struct vhdr_crc7 {
238 uint8_t crc7;
239};
240struct vhdr_meta {
241 uint64_t offset;
242 unsigned char thread;
243 unsigned short numberio;
244 unsigned long time_sec;
245 unsigned long time_usec;
246};
247
248struct group_run_stats {
249 unsigned long long max_run[2], min_run[2];
250 unsigned long long max_bw[2], min_bw[2];
251 unsigned long long io_kb[2];
252 unsigned long long agg[2];
253};
254
255/*
256 * What type of allocation to use for io buffers
257 */
258enum fio_memtype {
259 MEM_MALLOC = 0, /* ordinary malloc */
260 MEM_SHM, /* use shared memory segments */
261 MEM_SHMHUGE, /* use shared memory segments with huge pages */
262 MEM_MMAP, /* use anonynomous mmap */
263 MEM_MMAPHUGE, /* memory mapped huge file */
264};
265
266/*
267 * The type of object we are working on
268 */
269enum fio_filetype {
270 FIO_TYPE_FILE = 1, /* plain file */
271 FIO_TYPE_BD, /* block device */
272 FIO_TYPE_CHAR, /* character device */
273 FIO_TYPE_PIPE, /* pipe */
274};
275
276enum fio_ioengine_flags {
277 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
278 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
279 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
280 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
281 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
282 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
283 FIO_NOIO = 1 << 6, /* thread does only pseudo IO */
284 FIO_SIGQUIT = 1 << 7, /* needs SIGQUIT to exit */
285};
286
287enum fio_file_flags {
288 FIO_FILE_OPEN = 1 << 0, /* file is open */
289 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
290 FIO_FILE_EXTEND = 1 << 2, /* needs extend */
291 FIO_FILE_DONE = 1 << 3, /* io completed to this file */
292 FIO_SIZE_KNOWN = 1 << 4, /* size has been set */
293 FIO_FILE_HASHED = 1 << 5, /* file is on hash */
294};
295
296/*
297 * Each thread_data structure has a number of files associated with it,
298 * this structure holds state information for a single file.
299 */
300struct fio_file {
301 struct flist_head hash_list;
302 enum fio_filetype filetype;
303
304 /*
305 * A file may not be a file descriptor, let the io engine decide
306 */
307 union {
308 unsigned long file_data;
309 int fd;
310 };
311
312 /*
313 * filename and possible memory mapping
314 */
315 char *file_name;
316 void *mmap;
317 unsigned int major, minor;
318
319 /*
320 * size of the file, offset into file, and io size from that offset
321 */
322 unsigned long long real_file_size;
323 unsigned long long file_offset;
324 unsigned long long io_size;
325
326 unsigned long long last_pos;
327
328 /*
329 * if io is protected by a semaphore, this is set
330 */
331 struct fio_mutex *lock;
332 void *lock_owner;
333 unsigned int lock_batch;
334 enum fio_ddir lock_ddir;
335
336 /*
337 * block map for random io
338 */
339 unsigned int *file_map;
340 unsigned int num_maps;
341 unsigned int last_free_lookup;
342
343 int references;
344 enum fio_file_flags flags;
345};
346
347/*
348 * How many depth levels to log
349 */
350#define FIO_IO_U_MAP_NR 8
351#define FIO_IO_U_LAT_U_NR 10
352#define FIO_IO_U_LAT_M_NR 12
353
354struct thread_stat {
355 char *name;
356 char *verror;
357 int error;
358 int groupid;
359 pid_t pid;
360 char *description;
361 int members;
362
363 struct io_log *slat_log;
364 struct io_log *clat_log;
365 struct io_log *bw_log;
366
367 /*
368 * bandwidth and latency stats
369 */
370 struct io_stat clat_stat[2]; /* completion latency */
371 struct io_stat slat_stat[2]; /* submission latency */
372 struct io_stat bw_stat[2]; /* bandwidth stats */
373
374 unsigned long long stat_io_bytes[2];
375 struct timeval stat_sample_time[2];
376
377 /*
378 * fio system usage accounting
379 */
380 struct rusage ru_start;
381 struct rusage ru_end;
382 unsigned long usr_time;
383 unsigned long sys_time;
384 unsigned long ctx;
385 unsigned long minf, majf;
386
387 /*
388 * IO depth and latency stats
389 */
390 unsigned int io_u_map[FIO_IO_U_MAP_NR];
391 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
392 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
393 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
394 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
395 unsigned long total_io_u[2];
396 unsigned long short_io_u[2];
397 unsigned long total_submit;
398 unsigned long total_complete;
399
400 unsigned long long io_bytes[2];
401 unsigned long runtime[2];
402 unsigned long total_run_time;
403};
404
405struct bssplit {
406 unsigned int bs;
407 unsigned char perc;
408};
409
410struct thread_options {
411 int pad;
412 char *description;
413 char *name;
414 char *directory;
415 char *filename;
416 char *opendir;
417 char *ioengine;
418 enum td_ddir td_ddir;
419 unsigned int ddir_nr;
420 unsigned int iodepth;
421 unsigned int iodepth_low;
422 unsigned int iodepth_batch;
423 unsigned int iodepth_batch_complete;
424
425 unsigned long long size;
426 unsigned int fill_device;
427 unsigned long long file_size_low;
428 unsigned long long file_size_high;
429 unsigned long long start_offset;
430
431 unsigned int bs[2];
432 unsigned int min_bs[2];
433 unsigned int max_bs[2];
434 struct bssplit *bssplit;
435 unsigned int bssplit_nr;
436
437 unsigned int nr_files;
438 unsigned int open_files;
439 enum file_lock_mode file_lock_mode;
440 unsigned int lockfile_batch;
441
442 unsigned int odirect;
443 unsigned int invalidate_cache;
444 unsigned int create_serialize;
445 unsigned int create_fsync;
446 unsigned int end_fsync;
447 unsigned int sync_io;
448 unsigned int verify;
449 unsigned int do_verify;
450 unsigned int verifysort;
451 unsigned int verify_interval;
452 unsigned int verify_offset;
453 unsigned int verify_pattern;
454 unsigned int verify_pattern_bytes;
455 unsigned int verify_fatal;
456 unsigned int use_thread;
457 unsigned int unlink;
458 unsigned int do_disk_util;
459 unsigned int override_sync;
460 unsigned int rand_repeatable;
461 unsigned int write_lat_log;
462 unsigned int write_bw_log;
463 unsigned int norandommap;
464 unsigned int softrandommap;
465 unsigned int bs_unaligned;
466 unsigned int fsync_on_close;
467
468 unsigned int hugepage_size;
469 unsigned int rw_min_bs;
470 unsigned int thinktime;
471 unsigned int thinktime_spin;
472 unsigned int thinktime_blocks;
473 unsigned int fsync_blocks;
474 unsigned int start_delay;
475 unsigned long long timeout;
476 unsigned long long ramp_time;
477 unsigned int overwrite;
478 unsigned int bw_avg_time;
479 unsigned int loops;
480 unsigned long long zone_size;
481 unsigned long long zone_skip;
482 enum fio_memtype mem_type;
483
484 unsigned int stonewall;
485 unsigned int new_group;
486 unsigned int numjobs;
487 os_cpu_mask_t cpumask;
488 unsigned int cpumask_set;
489 unsigned int iolog;
490 unsigned int rwmixcycle;
491 unsigned int rwmix[2];
492 unsigned int nice;
493 unsigned int file_service_type;
494 unsigned int group_reporting;
495 unsigned int fadvise_hint;
496 unsigned int zero_buffers;
497 unsigned int refill_buffers;
498 unsigned int time_based;
499 unsigned int disable_clat;
500 unsigned int disable_slat;
501 unsigned int disable_bw;
502 unsigned int gtod_reduce;
503 unsigned int gtod_cpu;
504 unsigned int gtod_offload;
505
506 char *read_iolog_file;
507 char *write_iolog_file;
508 char *bw_log_file;
509 char *lat_log_file;
510
511 /*
512 * Pre-run and post-run shell
513 */
514 char *exec_prerun;
515 char *exec_postrun;
516
517 unsigned int rate;
518 unsigned int ratemin;
519 unsigned int ratecycle;
520 unsigned int rate_iops;
521 unsigned int rate_iops_min;
522
523 char *ioscheduler;
524
525 /*
526 * CPU "io" cycle burner
527 */
528 unsigned int cpuload;
529 unsigned int cpucycle;
530};
531
532#define FIO_VERROR_SIZE 128
533
534/*
535 * This describes a single thread/process executing a fio job.
536 */
537struct thread_data {
538 struct thread_options o;
539 char verror[FIO_VERROR_SIZE];
540 pthread_t thread;
541 int thread_number;
542 int groupid;
543 struct thread_stat ts;
544 struct fio_file **files;
545 unsigned int files_index;
546 unsigned int nr_open_files;
547 unsigned int nr_done_files;
548 unsigned int nr_normal_files;
549 union {
550 unsigned int next_file;
551 os_random_state_t next_file_state;
552 };
553 int error;
554 int done;
555 pid_t pid;
556 char *orig_buffer;
557 size_t orig_buffer_size;
558 volatile int terminate;
559 volatile int runstate;
560 unsigned int ioprio;
561 unsigned int ioprio_set;
562 unsigned int last_was_sync;
563
564 char *mmapfile;
565 int mmapfd;
566
567 void *iolog_buf;
568 FILE *iolog_f;
569
570 char *sysfs_root;
571
572 os_random_state_t bsrange_state;
573 os_random_state_t verify_state;
574
575 int shm_id;
576
577 /*
578 * IO engine hooks, contains everything needed to submit an io_u
579 * to any of the available IO engines.
580 */
581 struct ioengine_ops *io_ops;
582
583 /*
584 * Current IO depth and list of free and busy io_u's.
585 */
586 unsigned int cur_depth;
587 unsigned int io_u_queued;
588 struct flist_head io_u_freelist;
589 struct flist_head io_u_busylist;
590 struct flist_head io_u_requeues;
591
592 /*
593 * Rate state
594 */
595 unsigned long rate_usec_cycle;
596 long rate_pending_usleep;
597 unsigned long rate_bytes;
598 unsigned long rate_blocks;
599 struct timeval lastrate;
600
601 unsigned long long total_io_size;
602
603 unsigned long io_issues[2];
604 unsigned long long io_blocks[2];
605 unsigned long long io_bytes[2];
606 unsigned long long io_skip_bytes;
607 unsigned long long this_io_bytes[2];
608 unsigned long long zone_bytes;
609 struct fio_mutex *mutex;
610
611 /*
612 * State for random io, a bitmap of blocks done vs not done
613 */
614 os_random_state_t random_state;
615
616 struct timeval start; /* start of this loop */
617 struct timeval epoch; /* time job was started */
618 struct timeval rw_end[2];
619 struct timeval last_issue;
620 struct timeval tv_cache;
621 unsigned int tv_cache_nr;
622 unsigned int tv_cache_mask;
623 unsigned int rw_end_set[2];
624 unsigned int ramp_time_over;
625
626 /*
627 * read/write mixed workload state
628 */
629 os_random_state_t rwmix_state;
630 unsigned long rwmix_issues;
631 enum fio_ddir rwmix_ddir;
632 unsigned int ddir_nr;
633
634 /*
635 * IO history logs for verification. We use a tree for sorting,
636 * if we are overwriting. Otherwise just use a fifo.
637 */
638 struct rb_root io_hist_tree;
639 struct flist_head io_hist_list;
640
641 /*
642 * For IO replaying
643 */
644 struct flist_head io_log_list;
645
646 /*
647 * for fileservice, how often to switch to a new file
648 */
649 unsigned int file_service_nr;
650 unsigned int file_service_left;
651 struct fio_file *file_service_file;
652
653 /*
654 * For generating file sizes
655 */
656 os_random_state_t file_size_state;
657};
658
659/*
660 * roundrobin available files, or choose one at random, or do each one
661 * serially.
662 */
663enum {
664 FIO_FSERVICE_RANDOM = 1,
665 FIO_FSERVICE_RR = 2,
666 FIO_FSERVICE_SEQ = 3,
667};
668
669/*
670 * when should interactive ETA output be generated
671 */
672enum {
673 FIO_ETA_AUTO,
674 FIO_ETA_ALWAYS,
675 FIO_ETA_NEVER,
676};
677
678#define __td_verror(td, err, msg, func) \
679 do { \
680 if ((td)->error) \
681 break; \
682 int e = (err); \
683 (td)->error = e; \
684 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
685 } while (0)
686
687
688#define td_verror(td, err, func) \
689 __td_verror((td), (err), strerror((err)), (func))
690#define td_vmsg(td, err, msg, func) \
691 __td_verror((td), (err), (msg), (func))
692
693extern int exitall_on_terminate;
694extern int thread_number;
695extern int nr_process, nr_thread;
696extern int shm_id;
697extern int groupid;
698extern int terse_output;
699extern int temp_stall_ts;
700extern unsigned long long mlock_size;
701extern unsigned long page_mask, page_size;
702extern int read_only;
703extern int eta_print;
704extern unsigned long done_secs;
705extern char *job_section;
706extern int fio_gtod_offload;
707extern int fio_gtod_cpu;
708
709extern struct thread_data *threads;
710
711#define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
712#define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
713#define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
714#define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
715#define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
716
717static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
718{
719 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
720}
721
722#define BLOCKS_PER_MAP (8 * sizeof(int))
723#define TO_MAP_BLOCK(f, b) (b)
724#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
725#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
726
727#define MAX_JOBS (1024)
728
729static inline int should_fsync(struct thread_data *td)
730{
731 if (td->last_was_sync)
732 return 0;
733 if (td->o.odirect)
734 return 0;
735 if (td_write(td) || td_rw(td) || td->o.override_sync)
736 return 1;
737
738 return 0;
739}
740
741/*
742 * Disk utils as read in /sys/block/<dev>/stat
743 */
744struct disk_util_stat {
745 unsigned ios[2];
746 unsigned merges[2];
747 unsigned long long sectors[2];
748 unsigned ticks[2];
749 unsigned io_ticks;
750 unsigned time_in_queue;
751};
752
753/*
754 * Per-device disk util management
755 */
756struct disk_util {
757 struct flist_head list;
758 /* If this disk is a slave, hook it into the master's
759 * list using this head.
760 */
761 struct flist_head slavelist;
762
763 char *name;
764 char *sysfs_root;
765 char path[256];
766 int major, minor;
767
768 struct disk_util_stat dus;
769 struct disk_util_stat last_dus;
770
771 /* For software raids, this entry maintains pointers to the
772 * entries for the slave devices. The disk_util entries for
773 * the slaves devices should primarily be maintained through
774 * the disk_list list, i.e. for memory allocation and
775 * de-allocation, etc. Whereas this list should be used only
776 * for aggregating a software RAID's disk util figures.
777 */
778 struct flist_head slaves;
779
780 unsigned long msec;
781 struct timeval time;
782};
783
784#define DISK_UTIL_MSEC (250)
785
786/*
787 * Log exports
788 */
789enum file_log_act {
790 FIO_LOG_ADD_FILE,
791 FIO_LOG_OPEN_FILE,
792 FIO_LOG_CLOSE_FILE,
793 FIO_LOG_UNLINK_FILE,
794};
795
796extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
797extern void log_io_u(struct thread_data *, struct io_u *);
798extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
799extern int __must_check init_iolog(struct thread_data *td);
800extern void log_io_piece(struct thread_data *, struct io_u *);
801extern void queue_io_piece(struct thread_data *, struct io_piece *);
802extern void prune_io_piece_log(struct thread_data *);
803extern void write_iolog_close(struct thread_data *);
804
805/*
806 * Logging
807 */
808extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long);
809extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long);
810extern void add_bw_sample(struct thread_data *, enum fio_ddir, struct timeval *);
811extern void show_run_stats(void);
812extern void init_disk_util(struct thread_data *);
813extern void update_rusage_stat(struct thread_data *);
814extern void update_io_ticks(void);
815extern void setup_log(struct io_log **);
816extern void finish_log(struct thread_data *, struct io_log *, const char *);
817extern void finish_log_named(struct thread_data *, struct io_log *, const char *, const char *);
818extern void __finish_log(struct io_log *, const char *);
819extern struct io_log *agg_io_log[2];
820extern int write_bw_log;
821extern void add_agg_sample(unsigned long, enum fio_ddir);
822
823/*
824 * Time functions
825 */
826extern unsigned long long utime_since(struct timeval *, struct timeval *);
827extern unsigned long long utime_since_now(struct timeval *);
828extern unsigned long mtime_since(struct timeval *, struct timeval *);
829extern unsigned long mtime_since_now(struct timeval *);
830extern unsigned long time_since_now(struct timeval *);
831extern unsigned long mtime_since_genesis(void);
832extern void usec_spin(unsigned int);
833extern void usec_sleep(struct thread_data *, unsigned long);
834extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
835extern void fill_start_time(struct timeval *);
836extern void fio_gettime(struct timeval *, void *);
837extern void fio_gtod_init(void);
838extern void fio_gtod_update(void);
839extern void set_genesis_time(void);
840extern int ramp_time_over(struct thread_data *);
841extern int in_ramp_time(struct thread_data *);
842
843/*
844 * Init/option functions
845 */
846extern int __must_check parse_options(int, char **);
847extern int fio_options_parse(struct thread_data *, char **, int);
848extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
849extern void fio_fill_default_options(struct thread_data *);
850extern int fio_show_option_help(const char *);
851extern void fio_options_dup_and_init(struct option *);
852extern void options_mem_dupe(struct thread_data *);
853extern void options_mem_free(struct thread_data *);
854#define FIO_GETOPT_JOB 0x89988998
855#define FIO_NR_OPTIONS 128
856
857/*
858 * File setup/shutdown
859 */
860extern void close_files(struct thread_data *);
861extern void close_and_free_files(struct thread_data *);
862extern int __must_check setup_files(struct thread_data *);
863extern int __must_check open_files(struct thread_data *);
864extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
865extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
866extern int __must_check generic_close_file(struct thread_data *, struct fio_file *);
867extern int add_file(struct thread_data *, const char *);
868extern void get_file(struct fio_file *);
869extern int __must_check put_file(struct thread_data *, struct fio_file *);
870extern void lock_file(struct thread_data *, struct fio_file *, enum fio_ddir);
871extern void unlock_file(struct thread_data *, struct fio_file *);
872extern void unlock_file_all(struct thread_data *, struct fio_file *);
873extern int add_dir_files(struct thread_data *, const char *);
874extern int init_random_map(struct thread_data *);
875extern void dup_files(struct thread_data *, struct thread_data *);
876extern int get_fileno(struct thread_data *, const char *);
877extern void free_release_files(struct thread_data *);
878
879/*
880 * ETA/status stuff
881 */
882extern void print_thread_status(void);
883extern void print_status_init(int);
884
885/*
886 * disk util stuff
887 */
888#ifdef FIO_HAVE_DISK_UTIL
889extern void show_disk_util(void);
890extern void init_disk_util(struct thread_data *);
891extern void update_io_ticks(void);
892#else
893#define show_disk_util()
894#define init_disk_util(td)
895#define update_io_ticks()
896#endif
897
898/*
899 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
900 * will never back again. It may cycle between running/verififying/fsyncing.
901 * Once the thread reaches TD_EXITED, it is just waiting for the core to
902 * reap it.
903 */
904enum {
905 TD_NOT_CREATED = 0,
906 TD_CREATED,
907 TD_INITIALIZED,
908 TD_RAMP,
909 TD_RUNNING,
910 TD_VERIFYING,
911 TD_FSYNCING,
912 TD_EXITED,
913 TD_REAPED,
914};
915
916extern void td_set_runstate(struct thread_data *, int);
917
918/*
919 * Verify helpers
920 */
921extern void populate_verify_io_u(struct thread_data *, struct io_u *);
922extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
923extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
924
925/*
926 * Memory helpers
927 */
928extern int __must_check fio_pin_memory(void);
929extern void fio_unpin_memory(void);
930extern int __must_check allocate_io_mem(struct thread_data *);
931extern void free_io_mem(struct thread_data *);
932
933/*
934 * io unit handling
935 */
936#define queue_full(td) flist_empty(&(td)->io_u_freelist)
937extern struct io_u *__get_io_u(struct thread_data *);
938extern struct io_u *get_io_u(struct thread_data *);
939extern void put_io_u(struct thread_data *, struct io_u *);
940extern void requeue_io_u(struct thread_data *, struct io_u **);
941extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
942extern long __must_check io_u_queued_complete(struct thread_data *, int);
943extern void io_u_queued(struct thread_data *, struct io_u *);
944extern void io_u_log_error(struct thread_data *, struct io_u *);
945extern void io_u_mark_depth(struct thread_data *, unsigned int);
946extern void io_u_fill_buffer(struct thread_data *td, struct io_u *, unsigned int);
947void io_u_mark_complete(struct thread_data *, unsigned int);
948void io_u_mark_submit(struct thread_data *, unsigned int);
949
950/*
951 * Reset stats after ramp time completes
952 */
953extern void reset_all_stats(struct thread_data *);
954
955/*
956 * io engine entry points
957 */
958extern int __must_check td_io_init(struct thread_data *);
959extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
960extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
961extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
962extern int __must_check td_io_getevents(struct thread_data *, unsigned int, unsigned int, struct timespec *);
963extern int __must_check td_io_commit(struct thread_data *);
964extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
965extern int td_io_close_file(struct thread_data *, struct fio_file *);
966
967/*
968 * blktrace support
969 */
970#ifdef FIO_HAVE_BLKTRACE
971extern int is_blktrace(const char *);
972extern int load_blktrace(struct thread_data *, const char *);
973#endif
974
975struct ioengine_ops {
976 struct flist_head list;
977 char name[16];
978 int version;
979 int flags;
980 int (*setup)(struct thread_data *);
981 int (*init)(struct thread_data *);
982 int (*prep)(struct thread_data *, struct io_u *);
983 int (*queue)(struct thread_data *, struct io_u *);
984 int (*commit)(struct thread_data *);
985 int (*getevents)(struct thread_data *, unsigned int, unsigned int, struct timespec *);
986 struct io_u *(*event)(struct thread_data *, int);
987 int (*cancel)(struct thread_data *, struct io_u *);
988 void (*cleanup)(struct thread_data *);
989 int (*open_file)(struct thread_data *, struct fio_file *);
990 int (*close_file)(struct thread_data *, struct fio_file *);
991 void *data;
992 void *dlhandle;
993};
994
995#define FIO_IOOPS_VERSION 9
996
997extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
998extern void register_ioengine(struct ioengine_ops *);
999extern void unregister_ioengine(struct ioengine_ops *);
1000extern void close_ioengine(struct thread_data *);
1001
1002/*
1003 * Mark unused variables passed to ops functions as unused, to silence gcc
1004 */
1005#define fio_unused __attribute((__unused__))
1006#define fio_init __attribute__((constructor))
1007#define fio_exit __attribute__((destructor))
1008
1009#define for_each_td(td, i) \
1010 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
1011#define for_each_file(td, f, i) \
1012 if ((td)->files_index) \
1013 for ((i) = 0, (f) = (td)->files[0]; \
1014 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
1015 (i)++)
1016
1017#define fio_assert(td, cond) do { \
1018 if (!(cond)) { \
1019 int *__foo = NULL; \
1020 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
1021 (td)->runstate = TD_EXITED; \
1022 (td)->error = EFAULT; \
1023 *__foo = 0; \
1024 } \
1025} while (0)
1026
1027static inline void fio_file_reset(struct fio_file *f)
1028{
1029 f->last_free_lookup = 0;
1030 f->last_pos = f->file_offset;
1031}
1032
1033static inline void clear_error(struct thread_data *td)
1034{
1035 td->error = 0;
1036 td->verror[0] = '\0';
1037}
1038
1039#ifdef FIO_INC_DEBUG
1040static inline void dprint_io_u(struct io_u *io_u, const char *p)
1041{
1042 struct fio_file *f = io_u->file;
1043
1044 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
1045 (unsigned long long) io_u->offset,
1046 io_u->buflen, io_u->ddir);
1047 if (fio_debug & (1 << FD_IO)) {
1048 if (f)
1049 log_info("/%s", f->file_name);
1050
1051 log_info("\n");
1052 }
1053}
1054#else
1055#define dprint_io_u(io_u, p)
1056#endif
1057
1058static inline int fio_fill_issue_time(struct thread_data *td)
1059{
1060 if (td->o.read_iolog_file ||
1061 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
1062 return 1;
1063
1064 return 0;
1065}
1066
1067#endif