mmap engine: allow large files on 32-bit archs
[fio.git] / fio.h
CommitLineData
ebac4655
JA
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>
3c39a379
JA
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
6d6f031f 12#include <unistd.h>
34cfcdaf 13#include <string.h>
214e1eca 14#include <getopt.h>
cd14cc10 15#include <inttypes.h>
7101d9c2 16#include <assert.h>
ebac4655 17
317b95d0 18#include "compiler/compiler.h"
01743ee1 19#include "flist.h"
e2887563 20#include "fifo.h"
4b87898e 21#include "rbtree.h"
317b95d0
JA
22#include "arch/arch.h"
23#include "os/os.h"
07739b57 24#include "mutex.h"
a3d741fa
JA
25#include "log.h"
26#include "debug.h"
ebac4655 27
609342ff
DL
28#ifdef FIO_HAVE_GUASI
29#include <guasi.h>
30#endif
31
417f0068
JA
32#ifdef FIO_HAVE_SOLARISAIO
33#include <sys/asynch.h>
34#endif
35
1e97cce9
JA
36enum fio_ddir {
37 DDIR_READ = 0,
38 DDIR_WRITE,
39 DDIR_SYNC,
f29b25a3 40 DDIR_INVAL = -1,
1e97cce9
JA
41};
42
413dd459 43enum td_ddir {
b1ec1da6
JA
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,
7ff1c14b 49 TD_DDIR_RANDWRITE = TD_DDIR_WRITE | TD_DDIR_RAND,
b1ec1da6 50 TD_DDIR_RANDRW = TD_DDIR_RW | TD_DDIR_RAND,
413dd459
JA
51};
52
4d4e80f2
JA
53enum file_lock_mode {
54 FILE_LOCK_NONE,
55 FILE_LOCK_EXCLUSIVE,
56 FILE_LOCK_READWRITE,
57};
58
1c9e06ee
JA
59/*
60 * Use for maintaining statistics
61 */
ebac4655 62struct io_stat {
ebac4655
JA
63 unsigned long max_val;
64 unsigned long min_val;
65 unsigned long samples;
68704084
JA
66
67 double mean;
68 double S;
ebac4655
JA
69};
70
1c9e06ee
JA
71/*
72 * A single data sample
73 */
ebac4655
JA
74struct io_sample {
75 unsigned long time;
76 unsigned long val;
1e97cce9 77 enum fio_ddir ddir;
306ddc97 78 unsigned int bs;
ebac4655
JA
79};
80
1c9e06ee
JA
81/*
82 * Dynamically growing data sample log
83 */
ebac4655
JA
84struct io_log {
85 unsigned long nr_samples;
86 unsigned long max_samples;
87 struct io_sample *log;
88};
89
1c9e06ee
JA
90/*
91 * When logging io actions, this matches a single sent io_u
92 */
ebac4655 93struct io_piece {
4b87898e
JA
94 union {
95 struct rb_node rb_node;
01743ee1 96 struct flist_head list;
4b87898e 97 };
f29b25a3
JA
98 union {
99 int fileno;
100 struct fio_file *file;
101 };
ebac4655 102 unsigned long long offset;
a4f4fdd7 103 unsigned long len;
1e97cce9 104 enum fio_ddir ddir;
f29b25a3
JA
105 union {
106 unsigned long delay;
107 unsigned int file_action;
108 };
ebac4655
JA
109};
110
0c6e7517
JA
111enum {
112 IO_U_F_FREE = 1 << 0,
113 IO_U_F_FLIGHT = 1 << 1,
114};
115
36690c9b
JA
116struct thread_data;
117
ebac4655
JA
118/*
119 * The io unit
120 */
121struct io_u {
122 union {
123#ifdef FIO_HAVE_LIBAIO
124 struct iocb iocb;
125#endif
126#ifdef FIO_HAVE_POSIXAIO
127 struct aiocb aiocb;
128#endif
129#ifdef FIO_HAVE_SGIO
130 struct sg_io_hdr hdr;
a4f4fdd7 131#endif
609342ff
DL
132#ifdef FIO_HAVE_GUASI
133 guasi_req_t greq;
417f0068
JA
134#endif
135#ifdef FIO_HAVE_SOLARISAIO
136 aio_result_t resultp;
ebac4655 137#endif
ac893112 138 void *mmap_data;
ebac4655
JA
139 };
140 struct timeval start_time;
141 struct timeval issue_time;
142
1c9e06ee
JA
143 /*
144 * Allocated/set buffer and length
145 */
6b9cea23 146 void *buf;
a4f4fdd7 147 unsigned long buflen;
ebac4655
JA
148 unsigned long long offset;
149
1c9e06ee
JA
150 /*
151 * IO engine state, may be different from above when we get
152 * partial transfers / residual data counts
153 */
cec6b55d 154 void *xfer_buf;
a4f4fdd7 155 unsigned long xfer_buflen;
cec6b55d 156
ebac4655
JA
157 unsigned int resid;
158 unsigned int error;
159
1e97cce9 160 enum fio_ddir ddir;
ebac4655 161
dfd7bc2c
JA
162 /*
163 * io engine private data
164 */
165 union {
166 unsigned int index;
167 unsigned int seen;
556e831d 168 void *engine_data;
dfd7bc2c
JA
169 };
170
0c6e7517
JA
171 unsigned int flags;
172
53cdc686
JA
173 struct fio_file *file;
174
01743ee1 175 struct flist_head list;
d7762cf8
JA
176
177 /*
178 * Callback for io completion
179 */
36690c9b 180 int (*end_io)(struct thread_data *, struct io_u *);
ebac4655
JA
181};
182
36167d82
JA
183/*
184 * io_ops->queue() return values
185 */
186enum {
187 FIO_Q_COMPLETED = 0, /* completed sync */
188 FIO_Q_QUEUED = 1, /* queued, will complete async */
755200a3 189 FIO_Q_BUSY = 2, /* no more room, call ->commit() */
36167d82
JA
190};
191
ebac4655
JA
192#define FIO_HDR_MAGIC 0xf00baaef
193
194enum {
1c9e06ee
JA
195 VERIFY_NONE = 0, /* no verification */
196 VERIFY_MD5, /* md5 sum data blocks */
d77a7af3 197 VERIFY_CRC64, /* crc64 sum data blocks */
1c9e06ee 198 VERIFY_CRC32, /* crc32 sum data blocks */
bac39e0e 199 VERIFY_CRC32C, /* crc32c sum data blocks */
3845591f 200 VERIFY_CRC32C_INTEL, /* crc32c sum data blocks with hw */
969f7ed3 201 VERIFY_CRC16, /* crc16 sum data blocks */
1e154bdb 202 VERIFY_CRC7, /* crc7 sum data blocks */
cd14cc10
JA
203 VERIFY_SHA256, /* sha256 sum data blocks */
204 VERIFY_SHA512, /* sha512 sum data blocks */
7437ee87 205 VERIFY_META, /* block_num, timestamp etc. */
36690c9b 206 VERIFY_NULL, /* pretend to verify */
ebac4655
JA
207};
208
1c9e06ee 209/*
546dfd9f
JA
210 * A header structure associated with each checksummed data block. It is
211 * followed by a checksum specific header that contains the verification
212 * data.
1c9e06ee 213 */
ebac4655
JA
214struct verify_header {
215 unsigned int fio_magic;
216 unsigned int len;
217 unsigned int verify_type;
546dfd9f
JA
218};
219
220struct vhdr_md5 {
221 uint32_t md5_digest[16];
222};
223struct vhdr_sha512 {
224 uint8_t sha512[128];
225};
226struct vhdr_sha256 {
227 uint8_t sha256[128];
228};
229struct vhdr_crc64 {
230 uint64_t crc64;
231};
232struct vhdr_crc32 {
233 uint32_t crc32;
234};
235struct vhdr_crc16 {
236 uint16_t crc16;
237};
238struct vhdr_crc7 {
239 uint8_t crc7;
ebac4655 240};
7437ee87
SL
241struct vhdr_meta {
242 uint64_t offset;
243 unsigned char thread;
244 unsigned short numberio;
245 unsigned long time_sec;
246 unsigned long time_usec;
247};
ebac4655
JA
248
249struct group_run_stats {
9104f874
JA
250 unsigned long long max_run[2], min_run[2];
251 unsigned long long max_bw[2], min_bw[2];
e9b2a3fa 252 unsigned long long io_kb[2];
9104f874 253 unsigned long long agg[2];
ebac4655
JA
254};
255
e9c047a0
JA
256/*
257 * What type of allocation to use for io buffers
258 */
259enum fio_memtype {
260 MEM_MALLOC = 0, /* ordinary malloc */
261 MEM_SHM, /* use shared memory segments */
74b025b0 262 MEM_SHMHUGE, /* use shared memory segments with huge pages */
e9c047a0 263 MEM_MMAP, /* use anonynomous mmap */
d0bdaf49 264 MEM_MMAPHUGE, /* memory mapped huge file */
e9c047a0
JA
265};
266
267/*
268 * The type of object we are working on
269 */
270enum fio_filetype {
1c9e06ee
JA
271 FIO_TYPE_FILE = 1, /* plain file */
272 FIO_TYPE_BD, /* block device */
273 FIO_TYPE_CHAR, /* character device */
b5605e9d 274 FIO_TYPE_PIPE, /* pipe */
e9c047a0
JA
275};
276
2866c82d 277enum fio_ioengine_flags {
1c9e06ee 278 FIO_SYNCIO = 1 << 0, /* io engine has synchronous ->queue */
ba0fbe10
JA
279 FIO_RAWIO = 1 << 1, /* some sort of direct/raw io */
280 FIO_DISKLESSIO = 1 << 2, /* no disk involved */
0263882a 281 FIO_NOEXTEND = 1 << 3, /* engine can't extend file */
62d984e2 282 FIO_NODISKUTIL = 1 << 4, /* diskutil can't handle filename */
b67740d3 283 FIO_UNIDIR = 1 << 5, /* engine is uni-directional */
1f809d15 284 FIO_NOIO = 1 << 6, /* thread does only pseudo IO */
ad830ed7 285 FIO_SIGQUIT = 1 << 7, /* needs SIGQUIT to exit */
e9c047a0
JA
286};
287
f11bd94d 288enum fio_file_flags {
0ad920e7 289 FIO_FILE_OPEN = 1 << 0, /* file is open */
a9b7b305 290 FIO_FILE_CLOSING = 1 << 1, /* file being closed */
6ae1f57f
JA
291 FIO_FILE_EXTEND = 1 << 2, /* needs extend */
292 FIO_FILE_DONE = 1 << 3, /* io completed to this file */
293 FIO_SIZE_KNOWN = 1 << 4, /* size has been set */
392a3742 294 FIO_FILE_HASHED = 1 << 5, /* file is on hash */
f11bd94d
JA
295};
296
1c9e06ee
JA
297/*
298 * Each thread_data structure has a number of files associated with it,
299 * this structure holds state information for a single file.
300 */
53cdc686 301struct fio_file {
01743ee1 302 struct flist_head hash_list;
af52b345
JA
303 enum fio_filetype filetype;
304
53cdc686
JA
305 /*
306 * A file may not be a file descriptor, let the io engine decide
307 */
308 union {
309 unsigned long file_data;
e26f8f7a 310 int fd;
53cdc686 311 };
7bb48f84
JA
312
313 /*
314 * filename and possible memory mapping
315 */
1315a68a 316 char *file_name;
5e6c2067 317 unsigned int major, minor;
7bb48f84 318
ac893112
JA
319 void *mmap_ptr;
320 size_t mmap_sz;
321 off_t mmap_off;
322
7bb48f84
JA
323 /*
324 * size of the file, offset into file, and io size from that offset
325 */
53cdc686
JA
326 unsigned long long real_file_size;
327 unsigned long long file_offset;
7bb48f84
JA
328 unsigned long long io_size;
329
53cdc686
JA
330 unsigned long long last_pos;
331
29c1349f
JA
332 /*
333 * if io is protected by a semaphore, this is set
334 */
4d4e80f2
JA
335 struct fio_mutex *lock;
336 void *lock_owner;
337 unsigned int lock_batch;
338 enum fio_ddir lock_ddir;
29c1349f 339
1c9e06ee
JA
340 /*
341 * block map for random io
342 */
de605666 343 unsigned int *file_map;
53cdc686 344 unsigned int num_maps;
1c9e06ee 345 unsigned int last_free_lookup;
9b031dc8 346
0ad920e7 347 int references;
f11bd94d 348 enum fio_file_flags flags;
c97bd0fa
JA
349
350 struct disk_util *du;
53cdc686
JA
351};
352
b2560f3c
JA
353/*
354 * How many depth levels to log
355 */
356#define FIO_IO_U_MAP_NR 8
a9a18a3f
JA
357#define FIO_IO_U_LAT_U_NR 10
358#define FIO_IO_U_LAT_M_NR 12
b2560f3c 359
079ad09b 360struct thread_stat {
756867bd
JA
361 char *name;
362 char *verror;
363 int error;
364 int groupid;
365 pid_t pid;
366 char *description;
6586ee89 367 int members;
756867bd 368
079ad09b
JA
369 struct io_log *slat_log;
370 struct io_log *clat_log;
371 struct io_log *bw_log;
372
373 /*
374 * bandwidth and latency stats
375 */
376 struct io_stat clat_stat[2]; /* completion latency */
377 struct io_stat slat_stat[2]; /* submission latency */
378 struct io_stat bw_stat[2]; /* bandwidth stats */
379
380 unsigned long long stat_io_bytes[2];
381 struct timeval stat_sample_time[2];
382
383 /*
384 * fio system usage accounting
385 */
386 struct rusage ru_start;
387 struct rusage ru_end;
388 unsigned long usr_time;
389 unsigned long sys_time;
390 unsigned long ctx;
81887d5d 391 unsigned long minf, majf;
079ad09b 392
b2560f3c
JA
393 /*
394 * IO depth and latency stats
395 */
396 unsigned int io_u_map[FIO_IO_U_MAP_NR];
838bc709
JA
397 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
398 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
04a0feae
JA
399 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
400 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
b3605062 401 unsigned long total_io_u[2];
30061b97 402 unsigned long short_io_u[2];
838bc709
JA
403 unsigned long total_submit;
404 unsigned long total_complete;
756867bd
JA
405
406 unsigned long long io_bytes[2];
407 unsigned long runtime[2];
408 unsigned long total_run_time;
b2560f3c 409};
71619dc2 410
564ca972
JA
411struct bssplit {
412 unsigned int bs;
413 unsigned char perc;
414};
415
2dc1bbeb 416struct thread_options {
b1ec1da6 417 int pad;
61697c37 418 char *description;
b4692828 419 char *name;
ef899b63 420 char *directory;
13f8e2d2 421 char *filename;
2dc1bbeb 422 char *opendir;
09629a90 423 char *ioengine;
413dd459 424 enum td_ddir td_ddir;
211097b2 425 unsigned int ddir_nr;
2dc1bbeb
JA
426 unsigned int iodepth;
427 unsigned int iodepth_low;
428 unsigned int iodepth_batch;
4950421a 429 unsigned int iodepth_batch_complete;
2dc1bbeb
JA
430
431 unsigned long long size;
aa31f1f1 432 unsigned int fill_device;
2dc1bbeb
JA
433 unsigned long long file_size_low;
434 unsigned long long file_size_high;
435 unsigned long long start_offset;
436
437 unsigned int bs[2];
2b7a01d0 438 unsigned int ba[2];
2dc1bbeb
JA
439 unsigned int min_bs[2];
440 unsigned int max_bs[2];
720e84ad
JA
441 struct bssplit *bssplit[2];
442 unsigned int bssplit_nr[2];
2dc1bbeb
JA
443
444 unsigned int nr_files;
445 unsigned int open_files;
4d4e80f2 446 enum file_lock_mode file_lock_mode;
29c1349f 447 unsigned int lockfile_batch;
e9c047a0 448
9158d2f7
JA
449 unsigned int odirect;
450 unsigned int invalidate_cache;
451 unsigned int create_serialize;
452 unsigned int create_fsync;
814452bd 453 unsigned int create_on_open;
9158d2f7 454 unsigned int end_fsync;
afad68f7 455 unsigned int pre_read;
9158d2f7
JA
456 unsigned int sync_io;
457 unsigned int verify;
e84c73a8 458 unsigned int do_verify;
160b966d 459 unsigned int verifysort;
a59e170d
JA
460 unsigned int verify_interval;
461 unsigned int verify_offset;
90059d65
JA
462 unsigned int verify_pattern;
463 unsigned int verify_pattern_bytes;
a12a3b4d 464 unsigned int verify_fatal;
9158d2f7
JA
465 unsigned int use_thread;
466 unsigned int unlink;
467 unsigned int do_disk_util;
468 unsigned int override_sync;
469 unsigned int rand_repeatable;
470 unsigned int write_lat_log;
471 unsigned int write_bw_log;
bb8895e0 472 unsigned int norandommap;
2b386d25 473 unsigned int softrandommap;
690adba3 474 unsigned int bs_unaligned;
ebb1415f 475 unsigned int fsync_on_close;
e9c047a0 476
56bb17f2 477 unsigned int hugepage_size;
a00735e6 478 unsigned int rw_min_bs;
ebac4655 479 unsigned int thinktime;
48097d5c 480 unsigned int thinktime_spin;
9c1f7434 481 unsigned int thinktime_blocks;
ebac4655
JA
482 unsigned int fsync_blocks;
483 unsigned int start_delay;
0db26797 484 unsigned long long timeout;
721938ae 485 unsigned long long ramp_time;
ebac4655 486 unsigned int overwrite;
ebac4655 487 unsigned int bw_avg_time;
ebac4655 488 unsigned int loops;
20dc95c4
JA
489 unsigned long long zone_size;
490 unsigned long long zone_skip;
e9c047a0 491 enum fio_memtype mem_type;
2dc1bbeb 492
ebac4655 493 unsigned int stonewall;
b3d62a75 494 unsigned int new_group;
ebac4655 495 unsigned int numjobs;
ebac4655 496 os_cpu_mask_t cpumask;
375b2695 497 unsigned int cpumask_set;
aea47d44 498 unsigned int iolog;
a6ccc7be 499 unsigned int rwmixcycle;
e47f799f 500 unsigned int rwmix[2];
b6f4d880 501 unsigned int nice;
0aabe160 502 unsigned int file_service_type;
b2560f3c 503 unsigned int group_reporting;
d2f3ac35 504 unsigned int fadvise_hint;
e9459e5a 505 unsigned int zero_buffers;
5973cafb 506 unsigned int refill_buffers;
cf4464ca 507 unsigned int time_based;
9520ebb9
JA
508 unsigned int disable_clat;
509 unsigned int disable_slat;
510 unsigned int disable_bw;
993bf48b 511 unsigned int gtod_reduce;
be4ecfdf
JA
512 unsigned int gtod_cpu;
513 unsigned int gtod_offload;
aea47d44 514
076efc7c
JA
515 char *read_iolog_file;
516 char *write_iolog_file;
e3cedca7
JA
517 char *bw_log_file;
518 char *lat_log_file;
2dc1bbeb
JA
519
520 /*
521 * Pre-run and post-run shell
522 */
523 char *exec_prerun;
524 char *exec_postrun;
525
526 unsigned int rate;
527 unsigned int ratemin;
528 unsigned int ratecycle;
529 unsigned int rate_iops;
530 unsigned int rate_iops_min;
531
532 char *ioscheduler;
533
534 /*
535 * CPU "io" cycle burner
536 */
537 unsigned int cpuload;
538 unsigned int cpucycle;
539};
540
e4e33258
JA
541#define FIO_VERROR_SIZE 128
542
2dc1bbeb
JA
543/*
544 * This describes a single thread/process executing a fio job.
545 */
546struct thread_data {
547 struct thread_options o;
e4e33258 548 char verror[FIO_VERROR_SIZE];
2dc1bbeb
JA
549 pthread_t thread;
550 int thread_number;
551 int groupid;
552 struct thread_stat ts;
126d65c6 553 struct fio_file **files;
dd87b2c9 554 unsigned int files_size;
2dc1bbeb
JA
555 unsigned int files_index;
556 unsigned int nr_open_files;
1020a139 557 unsigned int nr_done_files;
2dc1bbeb
JA
558 unsigned int nr_normal_files;
559 union {
560 unsigned int next_file;
561 os_random_state_t next_file_state;
562 };
563 int error;
20e354ef 564 int done;
2dc1bbeb
JA
565 pid_t pid;
566 char *orig_buffer;
567 size_t orig_buffer_size;
568 volatile int terminate;
569 volatile int runstate;
570 unsigned int ioprio;
ac684785 571 unsigned int ioprio_set;
2dc1bbeb
JA
572 unsigned int last_was_sync;
573
574 char *mmapfile;
575 int mmapfd;
576
843a7413
JA
577 void *iolog_buf;
578 FILE *iolog_f;
ebac4655 579
da86774e 580 char *sysfs_root;
da86774e 581
5bfc35d7
JA
582 unsigned long rand_seeds[6];
583
6dfd46b9
JA
584 os_random_state_t bsrange_state;
585 os_random_state_t verify_state;
ebac4655
JA
586
587 int shm_id;
588
e9c047a0
JA
589 /*
590 * IO engine hooks, contains everything needed to submit an io_u
591 * to any of the available IO engines.
592 */
2866c82d 593 struct ioengine_ops *io_ops;
ebac4655 594
e9c047a0
JA
595 /*
596 * Current IO depth and list of free and busy io_u's.
597 */
ebac4655 598 unsigned int cur_depth;
d8005759 599 unsigned int io_u_queued;
01743ee1
JA
600 struct flist_head io_u_freelist;
601 struct flist_head io_u_busylist;
602 struct flist_head io_u_requeues;
ebac4655 603
e9c047a0
JA
604 /*
605 * Rate state
606 */
ebac4655
JA
607 unsigned long rate_usec_cycle;
608 long rate_pending_usleep;
609 unsigned long rate_bytes;
4e991c23 610 unsigned long rate_blocks;
ebac4655
JA
611 struct timeval lastrate;
612
ebac4655
JA
613 unsigned long long total_io_size;
614
755200a3 615 unsigned long io_issues[2];
9104f874
JA
616 unsigned long long io_blocks[2];
617 unsigned long long io_bytes[2];
48f5abd3 618 unsigned long long io_skip_bytes;
9104f874 619 unsigned long long this_io_bytes[2];
079ad09b 620 unsigned long long zone_bytes;
cdd18ad8 621 struct fio_mutex *mutex;
ebac4655 622
e9c047a0
JA
623 /*
624 * State for random io, a bitmap of blocks done vs not done
625 */
6dfd46b9 626 os_random_state_t random_state;
ebac4655 627
ebac4655
JA
628 struct timeval start; /* start of this loop */
629 struct timeval epoch; /* time job was started */
38d77cae 630 struct timeval rw_end[2];
a61eddec 631 struct timeval last_issue;
993bf48b
JA
632 struct timeval tv_cache;
633 unsigned int tv_cache_nr;
634 unsigned int tv_cache_mask;
38d77cae 635 unsigned int rw_end_set[2];
721938ae 636 unsigned int ramp_time_over;
ebac4655 637
e9c047a0
JA
638 /*
639 * read/write mixed workload state
640 */
6dfd46b9 641 os_random_state_t rwmix_state;
e4928662 642 unsigned long rwmix_issues;
e9c047a0 643 enum fio_ddir rwmix_ddir;
211097b2 644 unsigned int ddir_nr;
a6ccc7be 645
e9c047a0 646 /*
8de8f047
JA
647 * IO history logs for verification. We use a tree for sorting,
648 * if we are overwriting. Otherwise just use a fifo.
e9c047a0 649 */
4b87898e 650 struct rb_root io_hist_tree;
01743ee1 651 struct flist_head io_hist_list;
8de8f047
JA
652
653 /*
654 * For IO replaying
655 */
01743ee1 656 struct flist_head io_log_list;
433afcb4 657
1907dbc6
JA
658 /*
659 * for fileservice, how often to switch to a new file
660 */
661 unsigned int file_service_nr;
662 unsigned int file_service_left;
663 struct fio_file *file_service_file;
9c60ce64
JA
664
665 /*
666 * For generating file sizes
667 */
668 os_random_state_t file_size_state;
ebac4655
JA
669};
670
0aabe160 671/*
a086c257
JA
672 * roundrobin available files, or choose one at random, or do each one
673 * serially.
0aabe160
JA
674 */
675enum {
676 FIO_FSERVICE_RANDOM = 1,
677 FIO_FSERVICE_RR = 2,
a086c257 678 FIO_FSERVICE_SEQ = 3,
0aabe160
JA
679};
680
e592a06b
AC
681/*
682 * when should interactive ETA output be generated
683 */
684enum {
685 FIO_ETA_AUTO,
686 FIO_ETA_ALWAYS,
687 FIO_ETA_NEVER,
688};
689
e1161c32 690#define __td_verror(td, err, msg, func) \
ebac4655 691 do { \
19abcd3d
JA
692 if ((td)->error) \
693 break; \
ebac4655
JA
694 int e = (err); \
695 (td)->error = e; \
e1161c32 696 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
ebac4655
JA
697 } while (0)
698
b990b5c0 699
e1161c32
JA
700#define td_verror(td, err, func) \
701 __td_verror((td), (err), strerror((err)), (func))
702#define td_vmsg(td, err, msg, func) \
703 __td_verror((td), (err), (msg), (func))
b990b5c0 704
ebac4655
JA
705extern int exitall_on_terminate;
706extern int thread_number;
9cedf167 707extern int nr_process, nr_thread;
ebac4655
JA
708extern int shm_id;
709extern int groupid;
c6ae0a5b 710extern int terse_output;
53cdc686 711extern int temp_stall_ts;
1e97cce9 712extern unsigned long long mlock_size;
cfc99db7 713extern unsigned long page_mask, page_size;
4241ea8f 714extern int read_only;
e592a06b 715extern int eta_print;
d3eeeabc 716extern unsigned long done_secs;
01f06b63 717extern char *job_section;
be4ecfdf
JA
718extern int fio_gtod_offload;
719extern int fio_gtod_cpu;
ebac4655
JA
720
721extern struct thread_data *threads;
722
2dc1bbeb
JA
723#define td_read(td) ((td)->o.td_ddir & TD_DDIR_READ)
724#define td_write(td) ((td)->o.td_ddir & TD_DDIR_WRITE)
725#define td_rw(td) (((td)->o.td_ddir & TD_DDIR_RW) == TD_DDIR_RW)
726#define td_random(td) ((td)->o.td_ddir & TD_DDIR_RAND)
303032ae 727#define file_randommap(td, f) (!(td)->o.norandommap && (f)->file_map)
ebac4655 728
7101d9c2
JA
729static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
730{
731 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
732}
733
de605666 734#define BLOCKS_PER_MAP (8 * sizeof(int))
aec2de20
JA
735#define TO_MAP_BLOCK(f, b) (b)
736#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
737#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
ebac4655
JA
738
739#define MAX_JOBS (1024)
740
87dc1ab1
JA
741static inline int should_fsync(struct thread_data *td)
742{
743 if (td->last_was_sync)
744 return 0;
2dc1bbeb 745 if (td->o.odirect)
87dc1ab1 746 return 0;
2dc1bbeb 747 if (td_write(td) || td_rw(td) || td->o.override_sync)
87dc1ab1
JA
748 return 1;
749
750 return 0;
751}
752
1c9e06ee
JA
753/*
754 * Disk utils as read in /sys/block/<dev>/stat
755 */
ebac4655
JA
756struct disk_util_stat {
757 unsigned ios[2];
758 unsigned merges[2];
759 unsigned long long sectors[2];
760 unsigned ticks[2];
761 unsigned io_ticks;
762 unsigned time_in_queue;
763};
764
1c9e06ee
JA
765/*
766 * Per-device disk util management
767 */
ebac4655 768struct disk_util {
01743ee1 769 struct flist_head list;
67423013
ST
770 /* If this disk is a slave, hook it into the master's
771 * list using this head.
772 */
773 struct flist_head slavelist;
ebac4655
JA
774
775 char *name;
e11c410c 776 char *sysfs_root;
ebac4655 777 char path[256];
07e5b264 778 int major, minor;
ebac4655
JA
779
780 struct disk_util_stat dus;
781 struct disk_util_stat last_dus;
782
67423013
ST
783 /* For software raids, this entry maintains pointers to the
784 * entries for the slave devices. The disk_util entries for
785 * the slaves devices should primarily be maintained through
786 * the disk_list list, i.e. for memory allocation and
787 * de-allocation, etc. Whereas this list should be used only
788 * for aggregating a software RAID's disk util figures.
789 */
790 struct flist_head slaves;
791
ebac4655
JA
792 unsigned long msec;
793 struct timeval time;
c97bd0fa
JA
794
795 struct fio_mutex *lock;
796 unsigned long users;
ebac4655
JA
797};
798
c97bd0fa
JA
799static inline void disk_util_inc(struct disk_util *du)
800{
801 if (du) {
802 fio_mutex_down(du->lock);
803 du->users++;
804 fio_mutex_up(du->lock);
805 }
806}
807
808static inline void disk_util_dec(struct disk_util *du)
809{
810 if (du) {
811 fio_mutex_down(du->lock);
812 du->users--;
813 fio_mutex_up(du->lock);
814 }
815}
816
ebac4655
JA
817#define DISK_UTIL_MSEC (250)
818
6796209a
JA
819/*
820 * Log exports
821 */
f29b25a3
JA
822enum file_log_act {
823 FIO_LOG_ADD_FILE,
824 FIO_LOG_OPEN_FILE,
825 FIO_LOG_CLOSE_FILE,
f718273e 826 FIO_LOG_UNLINK_FILE,
f29b25a3
JA
827};
828
72cb971b 829extern int __must_check read_iolog_get(struct thread_data *, struct io_u *);
f29b25a3
JA
830extern void log_io_u(struct thread_data *, struct io_u *);
831extern void log_file(struct thread_data *, struct fio_file *, enum file_log_act);
72cb971b 832extern int __must_check init_iolog(struct thread_data *td);
6796209a 833extern void log_io_piece(struct thread_data *, struct io_u *);
691c8fb0 834extern void queue_io_piece(struct thread_data *, struct io_piece *);
6796209a
JA
835extern void prune_io_piece_log(struct thread_data *);
836extern void write_iolog_close(struct thread_data *);
837
838/*
839 * Logging
840 */
306ddc97
JA
841extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long,
842 unsigned int);
843extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long,
844 unsigned int);
845extern void add_bw_sample(struct thread_data *, enum fio_ddir, unsigned int,
846 struct timeval *);
6796209a
JA
847extern void show_run_stats(void);
848extern void init_disk_util(struct thread_data *);
849extern void update_rusage_stat(struct thread_data *);
850extern void update_io_ticks(void);
8914a9d8
JA
851extern void setup_log(struct io_log **);
852extern void finish_log(struct thread_data *, struct io_log *, const char *);
e3cedca7 853extern void finish_log_named(struct thread_data *, struct io_log *, const char *, const char *);
bb3884d8 854extern void __finish_log(struct io_log *, const char *);
bb3884d8
JA
855extern struct io_log *agg_io_log[2];
856extern int write_bw_log;
306ddc97 857extern void add_agg_sample(unsigned long, enum fio_ddir, unsigned int);
6796209a
JA
858
859/*
860 * Time functions
861 */
10927316
SL
862extern unsigned long long utime_since(struct timeval *, struct timeval *);
863extern unsigned long long utime_since_now(struct timeval *);
6796209a
JA
864extern unsigned long mtime_since(struct timeval *, struct timeval *);
865extern unsigned long mtime_since_now(struct timeval *);
866extern unsigned long time_since_now(struct timeval *);
263e529f 867extern unsigned long mtime_since_genesis(void);
6dd6f2cd 868extern void usec_spin(unsigned int);
6796209a 869extern void usec_sleep(struct thread_data *, unsigned long);
413dd459 870extern void rate_throttle(struct thread_data *, unsigned long, unsigned int);
6043c579 871extern void fill_start_time(struct timeval *);
02bcaa8c 872extern void fio_gettime(struct timeval *, void *);
be4ecfdf
JA
873extern void fio_gtod_init(void);
874extern void fio_gtod_update(void);
a2f77c9f 875extern void set_genesis_time(void);
721938ae 876extern int ramp_time_over(struct thread_data *);
b29ee5b3 877extern int in_ramp_time(struct thread_data *);
6796209a 878
8914a9d8 879/*
214e1eca 880 * Init/option functions
8914a9d8 881 */
72cb971b 882extern int __must_check parse_options(int, char **);
3b8b7135 883extern int fio_options_parse(struct thread_data *, char **, int);
214e1eca
JA
884extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
885extern void fio_fill_default_options(struct thread_data *);
886extern int fio_show_option_help(const char *);
887extern void fio_options_dup_and_init(struct option *);
d23bb327
JA
888extern void options_mem_dupe(struct thread_data *);
889extern void options_mem_free(struct thread_data *);
5bfc35d7 890extern void td_fill_rand_seeds(struct thread_data *);
214e1eca
JA
891#define FIO_GETOPT_JOB 0x89988998
892#define FIO_NR_OPTIONS 128
8914a9d8 893
53cdc686
JA
894/*
895 * File setup/shutdown
896 */
897extern void close_files(struct thread_data *);
24ffd2c2 898extern void close_and_free_files(struct thread_data *);
b2fdda43 899extern int __must_check setup_files(struct thread_data *);
b2fdda43 900extern int __must_check file_invalidate_cache(struct thread_data *, struct fio_file *);
b5af8293 901extern int __must_check generic_open_file(struct thread_data *, struct fio_file *);
6977bcd0 902extern int __must_check generic_close_file(struct thread_data *, struct fio_file *);
df9c26b1 903extern int __must_check generic_get_file_size(struct thread_data *, struct fio_file *);
afad68f7 904extern int __must_check pre_read_files(struct thread_data *);
f29b25a3 905extern int add_file(struct thread_data *, const char *);
0ad920e7 906extern void get_file(struct fio_file *);
6977bcd0 907extern int __must_check put_file(struct thread_data *, struct fio_file *);
4d4e80f2
JA
908extern void lock_file(struct thread_data *, struct fio_file *, enum fio_ddir);
909extern void unlock_file(struct thread_data *, struct fio_file *);
910extern void unlock_file_all(struct thread_data *, struct fio_file *);
bbf6b540 911extern int add_dir_files(struct thread_data *, const char *);
68727076 912extern int init_random_map(struct thread_data *);
cade3ef4 913extern void dup_files(struct thread_data *, struct thread_data *);
f29b25a3
JA
914extern int get_fileno(struct thread_data *, const char *);
915extern void free_release_files(struct thread_data *);
53cdc686 916
263e529f
JA
917/*
918 * ETA/status stuff
919 */
920extern void print_thread_status(void);
921extern void print_status_init(int);
922
9f8f2064
JA
923/*
924 * disk util stuff
925 */
926#ifdef FIO_HAVE_DISK_UTIL
927extern void show_disk_util(void);
9f8f2064
JA
928extern void init_disk_util(struct thread_data *);
929extern void update_io_ticks(void);
930#else
931#define show_disk_util()
9f8f2064
JA
932#define init_disk_util(td)
933#define update_io_ticks()
934#endif
935
263e529f
JA
936/*
937 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
938 * will never back again. It may cycle between running/verififying/fsyncing.
939 * Once the thread reaches TD_EXITED, it is just waiting for the core to
940 * reap it.
941 */
942enum {
943 TD_NOT_CREATED = 0,
944 TD_CREATED,
945 TD_INITIALIZED,
b29ee5b3 946 TD_RAMP,
263e529f 947 TD_RUNNING,
b0f65863 948 TD_PRE_READING,
263e529f
JA
949 TD_VERIFYING,
950 TD_FSYNCING,
951 TD_EXITED,
952 TD_REAPED,
953};
954
b29ee5b3
JA
955extern void td_set_runstate(struct thread_data *, int);
956
e29d1b70
JA
957/*
958 * Verify helpers
959 */
960extern void populate_verify_io_u(struct thread_data *, struct io_u *);
b2fdda43 961extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
36690c9b 962extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
e29d1b70 963
2f9ade3c
JA
964/*
965 * Memory helpers
966 */
b2fdda43 967extern int __must_check fio_pin_memory(void);
2f9ade3c 968extern void fio_unpin_memory(void);
b2fdda43 969extern int __must_check allocate_io_mem(struct thread_data *);
2f9ade3c
JA
970extern void free_io_mem(struct thread_data *);
971
10ba535a
JA
972/*
973 * io unit handling
974 */
01743ee1 975#define queue_full(td) flist_empty(&(td)->io_u_freelist)
10ba535a 976extern struct io_u *__get_io_u(struct thread_data *);
3d7c391d 977extern struct io_u *get_io_u(struct thread_data *);
10ba535a 978extern void put_io_u(struct thread_data *, struct io_u *);
755200a3 979extern void requeue_io_u(struct thread_data *, struct io_u **);
d7762cf8
JA
980extern long __must_check io_u_sync_complete(struct thread_data *, struct io_u *);
981extern long __must_check io_u_queued_complete(struct thread_data *, int);
7e77dd02 982extern void io_u_queued(struct thread_data *, struct io_u *);
5451792e 983extern void io_u_log_error(struct thread_data *, struct io_u *);
d8005759 984extern void io_u_mark_depth(struct thread_data *, unsigned int);
5973cafb 985extern void io_u_fill_buffer(struct thread_data *td, struct io_u *, unsigned int);
838bc709
JA
986void io_u_mark_complete(struct thread_data *, unsigned int);
987void io_u_mark_submit(struct thread_data *, unsigned int);
10ba535a 988
b29ee5b3
JA
989/*
990 * Reset stats after ramp time completes
991 */
992extern void reset_all_stats(struct thread_data *);
993
10ba535a
JA
994/*
995 * io engine entry points
996 */
b2fdda43
JA
997extern int __must_check td_io_init(struct thread_data *);
998extern int __must_check td_io_prep(struct thread_data *, struct io_u *);
999extern int __must_check td_io_queue(struct thread_data *, struct io_u *);
1000extern int __must_check td_io_sync(struct thread_data *, struct fio_file *);
e7d2e616 1001extern int __must_check td_io_getevents(struct thread_data *, unsigned int, unsigned int, struct timespec *);
b2fdda43 1002extern int __must_check td_io_commit(struct thread_data *);
b5af8293 1003extern int __must_check td_io_open_file(struct thread_data *, struct fio_file *);
6977bcd0 1004extern int td_io_close_file(struct thread_data *, struct fio_file *);
df9c26b1 1005extern int __must_check td_io_get_file_size(struct thread_data *, struct fio_file *);
10ba535a 1006
fb7b71a3
JA
1007/*
1008 * blktrace support
1009 */
5e62c22a 1010#ifdef FIO_HAVE_BLKTRACE
fb7b71a3
JA
1011extern int is_blktrace(const char *);
1012extern int load_blktrace(struct thread_data *, const char *);
5e62c22a 1013#endif
fb7b71a3 1014
2866c82d 1015struct ioengine_ops {
01743ee1 1016 struct flist_head list;
2866c82d
JA
1017 char name[16];
1018 int version;
1019 int flags;
ea2877a4 1020 int (*setup)(struct thread_data *);
2866c82d
JA
1021 int (*init)(struct thread_data *);
1022 int (*prep)(struct thread_data *, struct io_u *);
1023 int (*queue)(struct thread_data *, struct io_u *);
755200a3 1024 int (*commit)(struct thread_data *);
e7d2e616 1025 int (*getevents)(struct thread_data *, unsigned int, unsigned int, struct timespec *);
2866c82d
JA
1026 struct io_u *(*event)(struct thread_data *, int);
1027 int (*cancel)(struct thread_data *, struct io_u *);
1028 void (*cleanup)(struct thread_data *);
b5af8293 1029 int (*open_file)(struct thread_data *, struct fio_file *);
6977bcd0 1030 int (*close_file)(struct thread_data *, struct fio_file *);
df9c26b1 1031 int (*get_file_size)(struct thread_data *, struct fio_file *);
2866c82d
JA
1032 void *data;
1033 void *dlhandle;
1034};
1035
df9c26b1 1036#define FIO_IOOPS_VERSION 10
2866c82d 1037
b4692828 1038extern struct ioengine_ops *load_ioengine(struct thread_data *, const char *);
b2fdda43 1039extern void register_ioengine(struct ioengine_ops *);
5f350952 1040extern void unregister_ioengine(struct ioengine_ops *);
2866c82d
JA
1041extern void close_ioengine(struct thread_data *);
1042
1043/*
1044 * Mark unused variables passed to ops functions as unused, to silence gcc
1045 */
1046#define fio_unused __attribute((__unused__))
5f350952
JA
1047#define fio_init __attribute__((constructor))
1048#define fio_exit __attribute__((destructor))
2866c82d 1049
34572e28
JA
1050#define for_each_td(td, i) \
1051 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686 1052#define for_each_file(td, f, i) \
691c8fb0
JA
1053 if ((td)->files_index) \
1054 for ((i) = 0, (f) = (td)->files[0]; \
1055 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
1056 (i)++)
53cdc686 1057
0032bf9f
JA
1058#define fio_assert(td, cond) do { \
1059 if (!(cond)) { \
340fd243 1060 int *__foo = NULL; \
0032bf9f
JA
1061 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
1062 (td)->runstate = TD_EXITED; \
437c9b71 1063 (td)->error = EFAULT; \
340fd243 1064 *__foo = 0; \
0032bf9f
JA
1065 } \
1066} while (0)
1067
009bd847
JA
1068static inline void fio_file_reset(struct fio_file *f)
1069{
1070 f->last_free_lookup = 0;
1071 f->last_pos = f->file_offset;
eaf9b417
JA
1072 if (f->file_map)
1073 memset(f->file_map, 0, f->num_maps * sizeof(int));
009bd847
JA
1074}
1075
9bf27b45
JA
1076static inline void clear_error(struct thread_data *td)
1077{
1078 td->error = 0;
1079 td->verror[0] = '\0';
1080}
1081
79e48f72 1082#ifdef FIO_INC_DEBUG
2ba1c290
JA
1083static inline void dprint_io_u(struct io_u *io_u, const char *p)
1084{
1085 struct fio_file *f = io_u->file;
1086
1087 dprint(FD_IO, "%s: io_u %p: off=%llu/len=%lu/ddir=%d", p, io_u,
5921e80c
JA
1088 (unsigned long long) io_u->offset,
1089 io_u->buflen, io_u->ddir);
8172fe97
JA
1090 if (fio_debug & (1 << FD_IO)) {
1091 if (f)
1092 log_info("/%s", f->file_name);
1093
1094 log_info("\n");
1095 }
2ba1c290 1096}
79e48f72 1097#else
79e48f72
JA
1098#define dprint_io_u(io_u, p)
1099#endif
2ba1c290 1100
12d9d841
JA
1101static inline int fio_fill_issue_time(struct thread_data *td)
1102{
1103 if (td->o.read_iolog_file ||
1104 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
1105 return 1;
1106
1107 return 0;
1108}
1109
ebac4655 1110#endif