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