fix early termination of runs caused by ramp_time > runtime
[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 <inttypes.h>
15#include <assert.h>
16
17struct thread_data;
18
19#include "compiler/compiler.h"
20#include "flist.h"
21#include "fifo.h"
22#include "rbtree.h"
23#include "arch/arch.h"
24#include "os/os.h"
25#include "mutex.h"
26#include "log.h"
27#include "debug.h"
28#include "file.h"
29#include "io_ddir.h"
30#include "ioengine.h"
31#include "iolog.h"
32#include "helpers.h"
33#include "options.h"
34#include "profile.h"
35#include "time.h"
36#include "lib/getopt.h"
37#include "lib/rand.h"
38#include "server.h"
39#include "stat.h"
40
41#ifdef FIO_HAVE_GUASI
42#include <guasi.h>
43#endif
44
45#ifdef FIO_HAVE_SOLARISAIO
46#include <sys/asynch.h>
47#endif
48
49/*
50 * What type of allocation to use for io buffers
51 */
52enum fio_memtype {
53 MEM_MALLOC = 0, /* ordinary malloc */
54 MEM_SHM, /* use shared memory segments */
55 MEM_SHMHUGE, /* use shared memory segments with huge pages */
56 MEM_MMAP, /* use anonynomous mmap */
57 MEM_MMAPHUGE, /* memory mapped huge file */
58};
59
60/*
61 * offset generator types
62 */
63enum {
64 RW_SEQ_SEQ = 0,
65 RW_SEQ_IDENT,
66};
67
68/*
69 * What type of errors to continue on when continue_on_error is used
70 */
71enum error_type {
72 ERROR_TYPE_NONE = 0,
73 ERROR_TYPE_READ = 1 << 0,
74 ERROR_TYPE_WRITE = 1 << 1,
75 ERROR_TYPE_VERIFY = 1 << 2,
76 ERROR_TYPE_ANY = 0xffff,
77};
78
79struct bssplit {
80 unsigned int bs;
81 unsigned char perc;
82};
83
84struct thread_options {
85 int pad;
86 char *description;
87 char *name;
88 char *directory;
89 char *filename;
90 char *opendir;
91 char *ioengine;
92 enum td_ddir td_ddir;
93 unsigned int rw_seq;
94 unsigned int kb_base;
95 unsigned int ddir_seq_nr;
96 long ddir_seq_add;
97 unsigned int iodepth;
98 unsigned int iodepth_low;
99 unsigned int iodepth_batch;
100 unsigned int iodepth_batch_complete;
101
102 unsigned long long size;
103 unsigned int size_percent;
104 unsigned int fill_device;
105 unsigned long long file_size_low;
106 unsigned long long file_size_high;
107 unsigned long long start_offset;
108
109 unsigned int bs[2];
110 unsigned int ba[2];
111 unsigned int min_bs[2];
112 unsigned int max_bs[2];
113 struct bssplit *bssplit[2];
114 unsigned int bssplit_nr[2];
115
116 unsigned int nr_files;
117 unsigned int open_files;
118 enum file_lock_mode file_lock_mode;
119 unsigned int lockfile_batch;
120
121 unsigned int odirect;
122 unsigned int invalidate_cache;
123 unsigned int create_serialize;
124 unsigned int create_fsync;
125 unsigned int create_on_open;
126 unsigned int end_fsync;
127 unsigned int pre_read;
128 unsigned int sync_io;
129 unsigned int verify;
130 unsigned int do_verify;
131 unsigned int verifysort;
132 unsigned int verify_interval;
133 unsigned int verify_offset;
134 char verify_pattern[MAX_PATTERN_SIZE];
135 unsigned int verify_pattern_bytes;
136 unsigned int verify_fatal;
137 unsigned int verify_dump;
138 unsigned int verify_async;
139 unsigned long long verify_backlog;
140 unsigned int verify_batch;
141 unsigned int use_thread;
142 unsigned int unlink;
143 unsigned int do_disk_util;
144 unsigned int override_sync;
145 unsigned int rand_repeatable;
146 unsigned int use_os_rand;
147 unsigned int write_lat_log;
148 unsigned int write_bw_log;
149 unsigned int write_iops_log;
150 unsigned int log_avg_msec;
151 unsigned int norandommap;
152 unsigned int softrandommap;
153 unsigned int bs_unaligned;
154 unsigned int fsync_on_close;
155
156 unsigned int hugepage_size;
157 unsigned int rw_min_bs;
158 unsigned int thinktime;
159 unsigned int thinktime_spin;
160 unsigned int thinktime_blocks;
161 unsigned int fsync_blocks;
162 unsigned int fdatasync_blocks;
163 unsigned int barrier_blocks;
164 unsigned long long start_delay;
165 unsigned long long timeout;
166 unsigned long long ramp_time;
167 unsigned int overwrite;
168 unsigned int bw_avg_time;
169 unsigned int iops_avg_time;
170 unsigned int loops;
171 unsigned long long zone_size;
172 unsigned long long zone_skip;
173 enum fio_memtype mem_type;
174 unsigned int mem_align;
175
176 unsigned int stonewall;
177 unsigned int new_group;
178 unsigned int numjobs;
179 os_cpu_mask_t cpumask;
180 unsigned int cpumask_set;
181 os_cpu_mask_t verify_cpumask;
182 unsigned int verify_cpumask_set;
183 unsigned int iolog;
184 unsigned int rwmixcycle;
185 unsigned int rwmix[2];
186 unsigned int nice;
187 unsigned int file_service_type;
188 unsigned int group_reporting;
189 unsigned int fadvise_hint;
190 enum fio_fallocate_mode fallocate_mode;
191 unsigned int zero_buffers;
192 unsigned int refill_buffers;
193 unsigned int scramble_buffers;
194 unsigned int time_based;
195 unsigned int disable_lat;
196 unsigned int disable_clat;
197 unsigned int disable_slat;
198 unsigned int disable_bw;
199 unsigned int gtod_reduce;
200 unsigned int gtod_cpu;
201 unsigned int gtod_offload;
202 enum fio_cs clocksource;
203 unsigned int no_stall;
204 unsigned int trim_percentage;
205 unsigned int trim_batch;
206 unsigned int trim_zero;
207 unsigned long long trim_backlog;
208 unsigned int clat_percentiles;
209 unsigned int overwrite_plist;
210 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
211
212 char *read_iolog_file;
213 char *write_iolog_file;
214 char *bw_log_file;
215 char *lat_log_file;
216 char *iops_log_file;
217 char *replay_redirect;
218
219 /*
220 * Pre-run and post-run shell
221 */
222 char *exec_prerun;
223 char *exec_postrun;
224
225 unsigned int rate[2];
226 unsigned int ratemin[2];
227 unsigned int ratecycle;
228 unsigned int rate_iops[2];
229 unsigned int rate_iops_min[2];
230
231 char *ioscheduler;
232
233 /*
234 * CPU "io" cycle burner
235 */
236 unsigned int cpuload;
237 unsigned int cpucycle;
238
239 /*
240 * I/O Error handling
241 */
242 enum error_type continue_on_error;
243
244 /*
245 * Benchmark profile type
246 */
247 char *profile;
248
249 /*
250 * blkio cgroup support
251 */
252 char *cgroup;
253 unsigned int cgroup_weight;
254 unsigned int cgroup_nodelete;
255
256 unsigned int uid;
257 unsigned int gid;
258
259 unsigned int sync_file_range;
260};
261
262/*
263 * This describes a single thread/process executing a fio job.
264 */
265struct thread_data {
266 struct thread_options o;
267 void *eo;
268 char verror[FIO_VERROR_SIZE];
269 pthread_t thread;
270 int thread_number;
271 int groupid;
272 struct thread_stat ts;
273
274 struct io_log *slat_log;
275 struct io_log *clat_log;
276 struct io_log *lat_log;
277 struct io_log *bw_log;
278 struct io_log *iops_log;
279
280 uint64_t stat_io_bytes[2];
281 struct timeval bw_sample_time;
282
283 uint64_t stat_io_blocks[2];
284 struct timeval iops_sample_time;
285
286 struct rusage ru_start;
287 struct rusage ru_end;
288
289 struct fio_file **files;
290 unsigned int files_size;
291 unsigned int files_index;
292 unsigned int nr_open_files;
293 unsigned int nr_done_files;
294 unsigned int nr_normal_files;
295 union {
296 unsigned int next_file;
297 os_random_state_t next_file_state;
298 struct frand_state __next_file_state;
299 };
300 int error;
301 int done;
302 pid_t pid;
303 char *orig_buffer;
304 size_t orig_buffer_size;
305 volatile int terminate;
306 volatile int runstate;
307 unsigned int ioprio;
308 unsigned int ioprio_set;
309 unsigned int last_was_sync;
310 enum fio_ddir last_ddir;
311
312 char *mmapfile;
313 int mmapfd;
314
315 void *iolog_buf;
316 FILE *iolog_f;
317
318 char *sysfs_root;
319
320 unsigned long rand_seeds[8];
321
322 union {
323 os_random_state_t bsrange_state;
324 struct frand_state __bsrange_state;
325 };
326 union {
327 os_random_state_t verify_state;
328 struct frand_state __verify_state;
329 };
330 union {
331 os_random_state_t trim_state;
332 struct frand_state __trim_state;
333 };
334
335 struct frand_state buf_state;
336
337 unsigned int verify_batch;
338 unsigned int trim_batch;
339
340 int shm_id;
341
342 /*
343 * IO engine hooks, contains everything needed to submit an io_u
344 * to any of the available IO engines.
345 */
346 struct ioengine_ops *io_ops;
347
348 /*
349 * Current IO depth and list of free and busy io_u's.
350 */
351 unsigned int cur_depth;
352 unsigned int io_u_queued;
353 struct flist_head io_u_freelist;
354 struct flist_head io_u_busylist;
355 struct flist_head io_u_requeues;
356 pthread_mutex_t io_u_lock;
357 pthread_cond_t free_cond;
358
359 /*
360 * async verify offload
361 */
362 struct flist_head verify_list;
363 pthread_t *verify_threads;
364 unsigned int nr_verify_threads;
365 pthread_cond_t verify_cond;
366 int verify_thread_exit;
367
368 /*
369 * Rate state
370 */
371 unsigned long long rate_bps[2];
372 long rate_pending_usleep[2];
373 unsigned long rate_bytes[2];
374 unsigned long rate_blocks[2];
375 struct timeval lastrate[2];
376
377 unsigned long long total_io_size;
378 unsigned long long fill_device_size;
379
380 unsigned long io_issues[2];
381 unsigned long long io_blocks[2];
382 unsigned long long this_io_blocks[2];
383 unsigned long long io_bytes[2];
384 unsigned long long io_skip_bytes;
385 unsigned long long this_io_bytes[2];
386 unsigned long long zone_bytes;
387 struct fio_mutex *mutex;
388
389 /*
390 * State for random io, a bitmap of blocks done vs not done
391 */
392 union {
393 os_random_state_t random_state;
394 struct frand_state __random_state;
395 };
396
397 struct timeval start; /* start of this loop */
398 struct timeval epoch; /* time job was started */
399 struct timeval last_issue;
400 struct timeval tv_cache;
401 unsigned int tv_cache_nr;
402 unsigned int tv_cache_mask;
403 unsigned int ramp_time_over;
404
405 /*
406 * read/write mixed workload state
407 */
408 union {
409 os_random_state_t rwmix_state;
410 struct frand_state __rwmix_state;
411 };
412 unsigned long rwmix_issues;
413 enum fio_ddir rwmix_ddir;
414 unsigned int ddir_seq_nr;
415
416 /*
417 * IO history logs for verification. We use a tree for sorting,
418 * if we are overwriting. Otherwise just use a fifo.
419 */
420 struct rb_root io_hist_tree;
421 struct flist_head io_hist_list;
422 unsigned long io_hist_len;
423
424 /*
425 * For IO replaying
426 */
427 struct flist_head io_log_list;
428
429 /*
430 * For tracking/handling discards
431 */
432 struct flist_head trim_list;
433 unsigned long trim_entries;
434
435 /*
436 * for fileservice, how often to switch to a new file
437 */
438 unsigned int file_service_nr;
439 unsigned int file_service_left;
440 struct fio_file *file_service_file;
441
442 unsigned int sync_file_range_nr;
443
444 /*
445 * For generating file sizes
446 */
447 union {
448 os_random_state_t file_size_state;
449 struct frand_state __file_size_state;
450 };
451
452 /*
453 * Error counts
454 */
455 unsigned int total_err_count;
456 int first_error;
457
458 /*
459 * Can be overloaded by profiles
460 */
461 struct prof_io_ops prof_io_ops;
462 void *prof_data;
463};
464
465/*
466 * when should interactive ETA output be generated
467 */
468enum {
469 FIO_ETA_AUTO,
470 FIO_ETA_ALWAYS,
471 FIO_ETA_NEVER,
472};
473
474#define __td_verror(td, err, msg, func) \
475 do { \
476 if ((td)->error) \
477 break; \
478 int e = (err); \
479 (td)->error = e; \
480 if (!(td)->first_error) \
481 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
482 } while (0)
483
484
485#define td_clear_error(td) \
486 (td)->error = 0;
487#define td_verror(td, err, func) \
488 __td_verror((td), (err), strerror((err)), (func))
489#define td_vmsg(td, err, msg, func) \
490 __td_verror((td), (err), (msg), (func))
491
492#define __fio_stringify_1(x) #x
493#define __fio_stringify(x) __fio_stringify_1(x)
494
495extern int exitall_on_terminate;
496extern unsigned int thread_number;
497extern unsigned int nr_process, nr_thread;
498extern int shm_id;
499extern int groupid;
500extern int terse_output;
501extern int temp_stall_ts;
502extern unsigned long long mlock_size;
503extern unsigned long page_mask, page_size;
504extern int read_only;
505extern int eta_print;
506extern unsigned long done_secs;
507extern char *job_section;
508extern int fio_gtod_offload;
509extern int fio_gtod_cpu;
510extern enum fio_cs fio_clock_source;
511extern int warnings_fatal;
512extern int terse_version;
513extern int is_backend;
514extern int nr_clients;
515extern int log_syslog;
516extern const char fio_version_string[];
517extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
518
519extern struct thread_data *threads;
520
521static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
522{
523 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
524}
525
526#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
527#define TO_MAP_BLOCK(f, b) (b)
528#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
529#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
530
531#define REAL_MAX_JOBS 2048
532
533#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
534
535static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
536{
537 if (err == EILSEQ)
538 return ERROR_TYPE_VERIFY;
539 if (ddir == DDIR_READ)
540 return ERROR_TYPE_READ;
541 return ERROR_TYPE_WRITE;
542}
543
544static inline void update_error_count(struct thread_data *td, int err)
545{
546 td->total_err_count++;
547 if (td->total_err_count == 1)
548 td->first_error = err;
549}
550
551static inline int should_fsync(struct thread_data *td)
552{
553 if (td->last_was_sync)
554 return 0;
555 if (td->o.odirect)
556 return 0;
557 if (td_write(td) || td_rw(td) || td->o.override_sync)
558 return 1;
559
560 return 0;
561}
562
563/*
564 * Init/option functions
565 */
566extern int __must_check parse_options(int, char **);
567extern int parse_jobs_ini(char *, int, int);
568extern int parse_cmd_line(int, char **);
569extern int exec_run(void);
570extern void reset_fio_state(void);
571extern int fio_options_parse(struct thread_data *, char **, int);
572extern void fio_keywords_init(void);
573extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
574extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
575extern void fio_fill_default_options(struct thread_data *);
576extern int fio_show_option_help(const char *);
577extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
578extern void fio_options_dup_and_init(struct option *);
579extern void fio_options_mem_dupe(struct thread_data *);
580extern void options_mem_dupe(void *data, struct fio_option *options);
581extern void td_fill_rand_seeds(struct thread_data *);
582extern void add_job_opts(const char **);
583extern char *num2str(unsigned long, int, int, int);
584extern int ioengine_load(struct thread_data *);
585
586#define FIO_GETOPT_JOB 0x89000000
587#define FIO_GETOPT_IOENGINE 0x98000000
588#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
589
590/*
591 * ETA/status stuff
592 */
593extern void print_thread_status(void);
594extern void print_status_init(int);
595
596/*
597 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
598 * will never back again. It may cycle between running/verififying/fsyncing.
599 * Once the thread reaches TD_EXITED, it is just waiting for the core to
600 * reap it.
601 */
602enum {
603 TD_NOT_CREATED = 0,
604 TD_CREATED,
605 TD_INITIALIZED,
606 TD_RAMP,
607 TD_RUNNING,
608 TD_PRE_READING,
609 TD_VERIFYING,
610 TD_FSYNCING,
611 TD_EXITED,
612 TD_REAPED,
613};
614
615extern void td_set_runstate(struct thread_data *, int);
616#define TERMINATE_ALL (-1)
617extern void fio_terminate_threads(int);
618
619/*
620 * Memory helpers
621 */
622extern int __must_check fio_pin_memory(void);
623extern void fio_unpin_memory(void);
624extern int __must_check allocate_io_mem(struct thread_data *);
625extern void free_io_mem(struct thread_data *);
626
627/*
628 * Reset stats after ramp time completes
629 */
630extern void reset_all_stats(struct thread_data *);
631
632/*
633 * blktrace support
634 */
635#ifdef FIO_HAVE_BLKTRACE
636extern int is_blktrace(const char *);
637extern int load_blktrace(struct thread_data *, const char *);
638#endif
639
640/*
641 * Mark unused variables passed to ops functions as unused, to silence gcc
642 */
643#define fio_unused __attribute((__unused__))
644#define fio_init __attribute__((constructor))
645#define fio_exit __attribute__((destructor))
646
647#define for_each_td(td, i) \
648 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
649#define for_each_file(td, f, i) \
650 if ((td)->files_index) \
651 for ((i) = 0, (f) = (td)->files[0]; \
652 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
653 (i)++)
654
655#define fio_assert(td, cond) do { \
656 if (!(cond)) { \
657 int *__foo = NULL; \
658 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
659 td_set_runstate((td), TD_EXITED); \
660 (td)->error = EFAULT; \
661 *__foo = 0; \
662 } \
663} while (0)
664
665static inline int fio_fill_issue_time(struct thread_data *td)
666{
667 if (td->o.read_iolog_file ||
668 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
669 return 1;
670
671 return 0;
672}
673
674static inline int __should_check_rate(struct thread_data *td,
675 enum fio_ddir ddir)
676{
677 struct thread_options *o = &td->o;
678
679 /*
680 * If some rate setting was given, we need to check it
681 */
682 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
683 o->rate_iops_min[ddir])
684 return 1;
685
686 return 0;
687}
688
689static inline int should_check_rate(struct thread_data *td,
690 unsigned long *bytes_done)
691{
692 int ret = 0;
693
694 if (bytes_done[0])
695 ret |= __should_check_rate(td, 0);
696 if (bytes_done[1])
697 ret |= __should_check_rate(td, 1);
698
699 return ret;
700}
701
702static inline int is_power_of_2(unsigned int val)
703{
704 return (val != 0 && ((val & (val - 1)) == 0));
705}
706
707/*
708 * We currently only need to do locking if we have verifier threads
709 * accessing our internal structures too
710 */
711static inline void td_io_u_lock(struct thread_data *td)
712{
713 if (td->o.verify_async)
714 pthread_mutex_lock(&td->io_u_lock);
715}
716
717static inline void td_io_u_unlock(struct thread_data *td)
718{
719 if (td->o.verify_async)
720 pthread_mutex_unlock(&td->io_u_lock);
721}
722
723static inline void td_io_u_free_notify(struct thread_data *td)
724{
725 if (td->o.verify_async)
726 pthread_cond_signal(&td->free_cond);
727}
728
729extern const char *fio_get_arch_string(int);
730extern const char *fio_get_os_string(int);
731
732#endif