syslet v4 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
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 * Callback for io completion
150 */
151 int (*end_io)(struct io_u *);
152};
153
154/*
155 * io_ops->queue() return values
156 */
157enum {
158 FIO_Q_COMPLETED = 0, /* completed sync */
159 FIO_Q_QUEUED = 1, /* queued, will complete async */
160 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
161};
162
163#define FIO_HDR_MAGIC 0xf00baaef
164
165enum {
166 VERIFY_NONE = 0, /* no verification */
167 VERIFY_MD5, /* md5 sum data blocks */
168 VERIFY_CRC32, /* crc32 sum data blocks */
169};
170
171/*
172 * A header structure associated with each checksummed data block
173 */
174struct verify_header {
175 unsigned int fio_magic;
176 unsigned int len;
177 unsigned int verify_type;
178 union {
179 char md5_digest[MD5_HASH_WORDS * 4];
180 unsigned long crc32;
181 };
182};
183
184struct group_run_stats {
185 unsigned long long max_run[2], min_run[2];
186 unsigned long long max_bw[2], min_bw[2];
187 unsigned long long io_kb[2];
188 unsigned long long agg[2];
189};
190
191/*
192 * What type of allocation to use for io buffers
193 */
194enum fio_memtype {
195 MEM_MALLOC = 0, /* ordinary malloc */
196 MEM_SHM, /* use shared memory segments */
197 MEM_SHMHUGE, /* use shared memory segments with huge pages */
198 MEM_MMAP, /* use anonynomous mmap */
199 MEM_MMAPHUGE, /* memory mapped huge file */
200};
201
202/*
203 * The type of object we are working on
204 */
205enum fio_filetype {
206 FIO_TYPE_FILE = 1, /* plain file */
207 FIO_TYPE_BD, /* block device */
208 FIO_TYPE_CHAR, /* character device */
209};
210
211enum fio_ioengine_flags {
212 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
213 FIO_CPUIO = 1 << 1, /* cpu burner, doesn't do real io */
214 FIO_MMAPIO = 1 << 2, /* uses memory mapped io */
215 FIO_RAWIO = 1 << 3, /* some sort of direct/raw io */
216 FIO_NETIO = 1 << 4, /* networked io */
217 FIO_NULLIO = 1 << 5, /* no real data transfer (cpu/null) */
218};
219
220/*
221 * Each thread_data structure has a number of files associated with it,
222 * this structure holds state information for a single file.
223 */
224struct fio_file {
225 /*
226 * A file may not be a file descriptor, let the io engine decide
227 */
228 union {
229 unsigned long file_data;
230 int fd;
231 };
232 char *file_name;
233 void *mmap;
234 unsigned long long file_size;
235 unsigned long long real_file_size;
236 unsigned long long file_offset;
237 unsigned long long last_pos;
238 unsigned long long last_completed_pos;
239
240 /*
241 * block map for random io
242 */
243 unsigned long *file_map;
244 unsigned int num_maps;
245 unsigned int last_free_lookup;
246
247 unsigned int unlink;
248};
249
250struct thread_stat {
251 struct io_log *slat_log;
252 struct io_log *clat_log;
253 struct io_log *bw_log;
254
255 /*
256 * bandwidth and latency stats
257 */
258 struct io_stat clat_stat[2]; /* completion latency */
259 struct io_stat slat_stat[2]; /* submission latency */
260 struct io_stat bw_stat[2]; /* bandwidth stats */
261
262 unsigned long long stat_io_bytes[2];
263 struct timeval stat_sample_time[2];
264
265 /*
266 * fio system usage accounting
267 */
268 struct rusage ru_start;
269 struct rusage ru_end;
270 unsigned long usr_time;
271 unsigned long sys_time;
272 unsigned long ctx;
273};
274
275/*
276 * How many depth levels to log
277 */
278#define FIO_IO_U_MAP_NR 8
279#define FIO_IO_U_LAT_NR 12
280
281/*
282 * This describes a single thread/process executing a fio job.
283 */
284struct thread_data {
285 int pad;
286 char *description;
287 char *name;
288 char *directory;
289 char *filename;
290 char verror[80];
291 pthread_t thread;
292 int thread_number;
293 int groupid;
294 struct thread_stat ts;
295 enum fio_filetype filetype;
296 struct fio_file *files;
297 unsigned int nr_files;
298 unsigned int nr_uniq_files;
299 union {
300 unsigned int next_file;
301 os_random_state_t next_file_state;
302 };
303 int error;
304 pid_t pid;
305 char *orig_buffer;
306 size_t orig_buffer_size;
307 volatile int terminate;
308 volatile int runstate;
309 enum td_ddir td_ddir;
310 unsigned int ioprio;
311 unsigned int last_was_sync;
312
313 unsigned int odirect;
314 unsigned int invalidate_cache;
315 unsigned int create_serialize;
316 unsigned int create_fsync;
317 unsigned int end_fsync;
318 unsigned int sync_io;
319 unsigned int verify;
320 unsigned int use_thread;
321 unsigned int unlink;
322 unsigned int do_disk_util;
323 unsigned int override_sync;
324 unsigned int rand_repeatable;
325 unsigned int write_lat_log;
326 unsigned int write_bw_log;
327 unsigned int norandommap;
328 unsigned int bs_unaligned;
329
330 unsigned int bs[2];
331 unsigned int min_bs[2];
332 unsigned int max_bs[2];
333 unsigned int hugepage_size;
334 unsigned int rw_min_bs;
335 unsigned int thinktime;
336 unsigned int thinktime_spin;
337 unsigned int thinktime_blocks;
338 unsigned int fsync_blocks;
339 unsigned int start_delay;
340 unsigned long timeout;
341 unsigned int overwrite;
342 unsigned int bw_avg_time;
343 unsigned int loops;
344 unsigned long long zone_size;
345 unsigned long long zone_skip;
346 enum fio_memtype mem_type;
347 char *mmapfile;
348 int mmapfd;
349 unsigned int stonewall;
350 unsigned int numjobs;
351 unsigned int iodepth;
352 unsigned int iodepth_low;
353 os_cpu_mask_t cpumask;
354 unsigned int iolog;
355 unsigned int read_iolog;
356 unsigned int rwmixcycle;
357 unsigned int rwmixread;
358 unsigned int rwmixwrite;
359 unsigned int nice;
360 unsigned int file_service_type;
361
362 char *read_iolog_file;
363 char *write_iolog_file;
364 void *iolog_buf;
365 FILE *iolog_f;
366
367 char *sysfs_root;
368 char *ioscheduler;
369
370 os_random_state_t bsrange_state;
371 os_random_state_t verify_state;
372
373 int shm_id;
374
375 /*
376 * IO engine hooks, contains everything needed to submit an io_u
377 * to any of the available IO engines.
378 */
379 struct ioengine_ops *io_ops;
380
381 /*
382 * Current IO depth and list of free and busy io_u's.
383 */
384 unsigned int cur_depth;
385 unsigned int io_u_map[FIO_IO_U_MAP_NR];
386 unsigned int io_u_lat[FIO_IO_U_LAT_NR];
387 unsigned long total_io_u;
388 struct list_head io_u_freelist;
389 struct list_head io_u_busylist;
390 struct list_head io_u_requeues;
391
392 /*
393 * Rate state
394 */
395 unsigned int rate;
396 unsigned int ratemin;
397 unsigned int ratecycle;
398 unsigned long rate_usec_cycle;
399 long rate_pending_usleep;
400 unsigned long rate_bytes;
401 struct timeval lastrate;
402
403 unsigned long runtime[2]; /* msec */
404 unsigned long long io_size;
405 unsigned long long total_file_size;
406 unsigned long long start_offset;
407 unsigned long long total_io_size;
408
409 unsigned long io_issues[2];
410 unsigned long long io_blocks[2];
411 unsigned long long io_bytes[2];
412 unsigned long long this_io_bytes[2];
413 unsigned long long zone_bytes;
414 volatile int mutex;
415
416 /*
417 * State for random io, a bitmap of blocks done vs not done
418 */
419 os_random_state_t random_state;
420
421 /*
422 * CPU "io" cycle burner
423 */
424 unsigned int cpuload;
425 unsigned int cpucycle;
426
427 struct timeval start; /* start of this loop */
428 struct timeval epoch; /* time job was started */
429 struct timeval end_time;/* time job ended */
430
431 /*
432 * read/write mixed workload state
433 */
434 os_random_state_t rwmix_state;
435 struct timeval rwmix_switch;
436 enum fio_ddir rwmix_ddir;
437
438 /*
439 * Pre-run and post-run shell
440 */
441 char *exec_prerun;
442 char *exec_postrun;
443
444 /*
445 * IO historic logs
446 */
447 struct list_head io_hist_list;
448 struct list_head io_log_list;
449
450 /*
451 * timeout handling
452 */
453 struct timeval timeout_end;
454 struct itimerval timer;
455};
456
457/*
458 * roundrobin available files, or choose one at random.
459 */
460enum {
461 FIO_FSERVICE_RANDOM = 1,
462 FIO_FSERVICE_RR = 2,
463};
464
465/*
466 * 30 second per-io_u timeout, with 5 second intervals to avoid resetting
467 * the timer on each queue operation.
468 */
469#define IO_U_TIMEOUT_INC 5
470#define IO_U_TIMEOUT 30
471
472#define __td_verror(td, err, msg, func) \
473 do { \
474 if ((td)->error) \
475 break; \
476 int e = (err); \
477 (td)->error = e; \
478 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
479 } while (0)
480
481
482#define td_verror(td, err, func) \
483 __td_verror((td), (err), strerror((err)), (func))
484#define td_vmsg(td, err, msg, func) \
485 __td_verror((td), (err), (msg), (func))
486
487extern int exitall_on_terminate;
488extern int thread_number;
489extern int shm_id;
490extern int groupid;
491extern int terse_output;
492extern FILE *f_out;
493extern FILE *f_err;
494extern int temp_stall_ts;
495extern unsigned long long mlock_size;
496
497extern struct thread_data *threads;
498
499#define td_read(td) ((td)->td_ddir & TD_DDIR_READ)
500#define td_write(td) ((td)->td_ddir & TD_DDIR_WRITE)
501#define td_rw(td) (((td)->td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
502#define td_random(td) ((td)->td_ddir & TD_DDIR_RAND)
503
504#define BLOCKS_PER_MAP (8 * sizeof(long))
505#define TO_MAP_BLOCK(td, f, b) ((b) - ((f)->file_offset / (td)->rw_min_bs))
506#define RAND_MAP_IDX(td, f, b) (TO_MAP_BLOCK(td, f, b) / BLOCKS_PER_MAP)
507#define RAND_MAP_BIT(td, f, b) (TO_MAP_BLOCK(td, f, b) & (BLOCKS_PER_MAP - 1))
508
509#define MAX_JOBS (1024)
510
511static inline int should_fsync(struct thread_data *td)
512{
513 if (td->last_was_sync)
514 return 0;
515 if (td->odirect)
516 return 0;
517 if (td_write(td) || td_rw(td) || td->override_sync)
518 return 1;
519
520 return 0;
521}
522
523/*
524 * Disk utils as read in /sys/block/<dev>/stat
525 */
526struct disk_util_stat {
527 unsigned ios[2];
528 unsigned merges[2];
529 unsigned long long sectors[2];
530 unsigned ticks[2];
531 unsigned io_ticks;
532 unsigned time_in_queue;
533};
534
535/*
536 * Per-device disk util management
537 */
538struct disk_util {
539 struct list_head list;
540
541 char *name;
542 char path[256];
543 dev_t dev;
544
545 struct disk_util_stat dus;
546 struct disk_util_stat last_dus;
547
548 unsigned long msec;
549 struct timeval time;
550};
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 *);
667extern long __must_check io_u_queued_complete(struct thread_data *, int);
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