projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Better support for setting up and removing private cgroups
[fio.git] / fio.h
... / ...
CommitLineData
1#ifndef FIO_H
2#define FIO_H
3
4#include <sched.h>
5#include <limits.h>
6#include <pthread.h>
7#include <sys/time.h>
8#include <sys/resource.h>
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
12#include <unistd.h>
13#include <string.h>
14#include <getopt.h>
15#include <inttypes.h>
16#include <assert.h>
17
18#include "compiler/compiler.h"
19#include "flist.h"
20#include "fifo.h"
21#include "rbtree.h"
22#include "arch/arch.h"
23#include "os/os.h"
24#include "mutex.h"
25#include "log.h"
26#include "debug.h"
27#include "file.h"
28#include "io_ddir.h"
29#include "ioengine.h"
30#include "iolog.h"
31#include "profiles.h"
32#include "helpers.h"
33
34#ifdef FIO_HAVE_GUASI
35#include <guasi.h>
36#endif
37
38#ifdef FIO_HAVE_SOLARISAIO
39#include <sys/asynch.h>
40#endif
41
42struct group_run_stats {
43 unsigned long long max_run[2], min_run[2];
44 unsigned long long max_bw[2], min_bw[2];
45 unsigned long long io_kb[2];
46 unsigned long long agg[2];
47 unsigned int kb_base;
48};
49
50/*
51 * What type of allocation to use for io buffers
52 */
53enum fio_memtype {
54 MEM_MALLOC = 0, /* ordinary malloc */
55 MEM_SHM, /* use shared memory segments */
56 MEM_SHMHUGE, /* use shared memory segments with huge pages */
57 MEM_MMAP, /* use anonynomous mmap */
58 MEM_MMAPHUGE, /* memory mapped huge file */
59};
60
61/*
62 * How many depth levels to log
63 */
64#define FIO_IO_U_MAP_NR 8
65#define FIO_IO_U_LAT_U_NR 10
66#define FIO_IO_U_LAT_M_NR 12
67
68#define MAX_PATTERN_SIZE 512
69
70struct thread_stat {
71 char *name;
72 char *verror;
73 int error;
74 int groupid;
75 pid_t pid;
76 char *description;
77 int members;
78
79 struct io_log *slat_log;
80 struct io_log *clat_log;
81 struct io_log *bw_log;
82
83 /*
84 * bandwidth and latency stats
85 */
86 struct io_stat clat_stat[2]; /* completion latency */
87 struct io_stat slat_stat[2]; /* submission latency */
88 struct io_stat bw_stat[2]; /* bandwidth stats */
89
90 unsigned long long stat_io_bytes[2];
91 struct timeval stat_sample_time[2];
92
93 /*
94 * fio system usage accounting
95 */
96 struct rusage ru_start;
97 struct rusage ru_end;
98 unsigned long usr_time;
99 unsigned long sys_time;
100 unsigned long ctx;
101 unsigned long minf, majf;
102
103 /*
104 * IO depth and latency stats
105 */
106 unsigned int io_u_map[FIO_IO_U_MAP_NR];
107 unsigned int io_u_submit[FIO_IO_U_MAP_NR];
108 unsigned int io_u_complete[FIO_IO_U_MAP_NR];
109 unsigned int io_u_lat_u[FIO_IO_U_LAT_U_NR];
110 unsigned int io_u_lat_m[FIO_IO_U_LAT_M_NR];
111 unsigned long total_io_u[2];
112 unsigned long short_io_u[2];
113 unsigned long total_submit;
114 unsigned long total_complete;
115
116 unsigned long long io_bytes[2];
117 unsigned long runtime[2];
118 unsigned long total_run_time;
119
120 /*
121 * IO Error related stats
122 */
123 unsigned continue_on_error;
124 unsigned long total_err_count;
125 int first_error;
126
127 unsigned int kb_base;
128};
129
130struct bssplit {
131 unsigned int bs;
132 unsigned char perc;
133};
134
135struct thread_options {
136 int pad;
137 char *description;
138 char *name;
139 char *directory;
140 char *filename;
141 char *opendir;
142 char *ioengine;
143 enum td_ddir td_ddir;
144 unsigned int kb_base;
145 unsigned int ddir_nr;
146 unsigned int iodepth;
147 unsigned int iodepth_low;
148 unsigned int iodepth_batch;
149 unsigned int iodepth_batch_complete;
150
151 unsigned long long size;
152 unsigned int fill_device;
153 unsigned long long file_size_low;
154 unsigned long long file_size_high;
155 unsigned long long start_offset;
156
157 unsigned int bs[2];
158 unsigned int ba[2];
159 unsigned int min_bs[2];
160 unsigned int max_bs[2];
161 struct bssplit *bssplit[2];
162 unsigned int bssplit_nr[2];
163
164 unsigned int nr_files;
165 unsigned int open_files;
166 enum file_lock_mode file_lock_mode;
167 unsigned int lockfile_batch;
168
169 unsigned int odirect;
170 unsigned int invalidate_cache;
171 unsigned int create_serialize;
172 unsigned int create_fsync;
173 unsigned int create_on_open;
174 unsigned int end_fsync;
175 unsigned int pre_read;
176 unsigned int sync_io;
177 unsigned int verify;
178 unsigned int do_verify;
179 unsigned int verifysort;
180 unsigned int verify_interval;
181 unsigned int verify_offset;
182 char verify_pattern[MAX_PATTERN_SIZE];
183 unsigned int verify_pattern_bytes;
184 unsigned int verify_fatal;
185 unsigned int verify_async;
186 unsigned int use_thread;
187 unsigned int unlink;
188 unsigned int do_disk_util;
189 unsigned int override_sync;
190 unsigned int rand_repeatable;
191 unsigned int write_lat_log;
192 unsigned int write_bw_log;
193 unsigned int norandommap;
194 unsigned int softrandommap;
195 unsigned int bs_unaligned;
196 unsigned int fsync_on_close;
197
198 unsigned int hugepage_size;
199 unsigned int rw_min_bs;
200 unsigned int thinktime;
201 unsigned int thinktime_spin;
202 unsigned int thinktime_blocks;
203 unsigned int fsync_blocks;
204 unsigned int fdatasync_blocks;
205 unsigned int start_delay;
206 unsigned long long timeout;
207 unsigned long long ramp_time;
208 unsigned int overwrite;
209 unsigned int bw_avg_time;
210 unsigned int loops;
211 unsigned long long zone_size;
212 unsigned long long zone_skip;
213 enum fio_memtype mem_type;
214 unsigned int mem_align;
215
216 unsigned int stonewall;
217 unsigned int new_group;
218 unsigned int numjobs;
219 os_cpu_mask_t cpumask;
220 unsigned int cpumask_set;
221 os_cpu_mask_t verify_cpumask;
222 unsigned int verify_cpumask_set;
223 unsigned int iolog;
224 unsigned int rwmixcycle;
225 unsigned int rwmix[2];
226 unsigned int nice;
227 unsigned int file_service_type;
228 unsigned int group_reporting;
229 unsigned int fadvise_hint;
230 unsigned int zero_buffers;
231 unsigned int refill_buffers;
232 unsigned int time_based;
233 unsigned int disable_clat;
234 unsigned int disable_slat;
235 unsigned int disable_bw;
236 unsigned int gtod_reduce;
237 unsigned int gtod_cpu;
238 unsigned int gtod_offload;
239
240 char *read_iolog_file;
241 char *write_iolog_file;
242 char *bw_log_file;
243 char *lat_log_file;
244
245 /*
246 * Pre-run and post-run shell
247 */
248 char *exec_prerun;
249 char *exec_postrun;
250
251 unsigned int rate[2];
252 unsigned int ratemin[2];
253 unsigned int ratecycle;
254 unsigned int rate_iops[2];
255 unsigned int rate_iops_min[2];
256
257 char *ioscheduler;
258
259 /*
260 * CPU "io" cycle burner
261 */
262 unsigned int cpuload;
263 unsigned int cpucycle;
264
265 /*
266 * I/O Error handling
267 */
268 unsigned int continue_on_error;
269
270 /*
271 * Benchmark profile type
272 */
273 unsigned int profile;
274
275 /*
276 * blkio cgroup support
277 */
278 char *cgroup_root;
279 char *cgroup;
280 unsigned int cgroup_weight;
281};
282
283#define FIO_VERROR_SIZE 128
284
285/*
286 * This describes a single thread/process executing a fio job.
287 */
288struct thread_data {
289 struct thread_options o;
290 char verror[FIO_VERROR_SIZE];
291 pthread_t thread;
292 int thread_number;
293 int groupid;
294 struct thread_stat ts;
295 struct fio_file **files;
296 unsigned int files_size;
297 unsigned int files_index;
298 unsigned int nr_open_files;
299 unsigned int nr_done_files;
300 unsigned int nr_normal_files;
301 union {
302 unsigned int next_file;
303 os_random_state_t next_file_state;
304 };
305 int error;
306 int done;
307 pid_t pid;
308 char *orig_buffer;
309 size_t orig_buffer_size;
310 volatile int terminate;
311 volatile int runstate;
312 unsigned int ioprio;
313 unsigned int ioprio_set;
314 unsigned int last_was_sync;
315
316 char *mmapfile;
317 int mmapfd;
318
319 void *iolog_buf;
320 FILE *iolog_f;
321
322 char *sysfs_root;
323
324 unsigned long rand_seeds[6];
325
326 os_random_state_t bsrange_state;
327 os_random_state_t verify_state;
328
329 int shm_id;
330
331 /*
332 * IO engine hooks, contains everything needed to submit an io_u
333 * to any of the available IO engines.
334 */
335 struct ioengine_ops *io_ops;
336
337 /*
338 * Current IO depth and list of free and busy io_u's.
339 */
340 unsigned int cur_depth;
341 unsigned int io_u_queued;
342 struct flist_head io_u_freelist;
343 struct flist_head io_u_busylist;
344 struct flist_head io_u_requeues;
345 pthread_mutex_t io_u_lock;
346 pthread_cond_t free_cond;
347
348 /*
349 * async verify offload
350 */
351 struct flist_head verify_list;
352 pthread_t *verify_threads;
353 unsigned int nr_verify_threads;
354 pthread_cond_t verify_cond;
355 int verify_thread_exit;
356
357 /*
358 * Rate state
359 */
360 unsigned long rate_usec_cycle[2];
361 long rate_pending_usleep[2];
362 unsigned long rate_bytes[2];
363 unsigned long rate_blocks[2];
364 struct timeval lastrate[2];
365
366 unsigned long long total_io_size;
367
368 unsigned long io_issues[2];
369 unsigned long long io_blocks[2];
370 unsigned long long io_bytes[2];
371 unsigned long long io_skip_bytes;
372 unsigned long long this_io_bytes[2];
373 unsigned long long zone_bytes;
374 struct fio_mutex *mutex;
375
376 /*
377 * State for random io, a bitmap of blocks done vs not done
378 */
379 os_random_state_t random_state;
380
381 struct timeval start; /* start of this loop */
382 struct timeval epoch; /* time job was started */
383 struct timeval last_issue;
384 struct timeval tv_cache;
385 unsigned int tv_cache_nr;
386 unsigned int tv_cache_mask;
387 unsigned int ramp_time_over;
388
389 /*
390 * read/write mixed workload state
391 */
392 os_random_state_t rwmix_state;
393 unsigned long rwmix_issues;
394 enum fio_ddir rwmix_ddir;
395 unsigned int ddir_nr;
396
397 /*
398 * IO history logs for verification. We use a tree for sorting,
399 * if we are overwriting. Otherwise just use a fifo.
400 */
401 struct rb_root io_hist_tree;
402 struct flist_head io_hist_list;
403
404 /*
405 * For IO replaying
406 */
407 struct flist_head io_log_list;
408
409 /*
410 * for fileservice, how often to switch to a new file
411 */
412 unsigned int file_service_nr;
413 unsigned int file_service_left;
414 struct fio_file *file_service_file;
415
416 /*
417 * For generating file sizes
418 */
419 os_random_state_t file_size_state;
420
421 /*
422 * Error counts
423 */
424 unsigned int total_err_count;
425 int first_error;
426};
427
428/*
429 * when should interactive ETA output be generated
430 */
431enum {
432 FIO_ETA_AUTO,
433 FIO_ETA_ALWAYS,
434 FIO_ETA_NEVER,
435};
436
437#define __td_verror(td, err, msg, func) \
438 do { \
439 if ((td)->error) \
440 break; \
441 int e = (err); \
442 (td)->error = e; \
443 if (!(td)->first_error) \
444 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
445 } while (0)
446
447
448#define td_clear_error(td) \
449 (td)->error = 0;
450#define td_verror(td, err, func) \
451 __td_verror((td), (err), strerror((err)), (func))
452#define td_vmsg(td, err, msg, func) \
453 __td_verror((td), (err), (msg), (func))
454
455extern int exitall_on_terminate;
456extern int thread_number;
457extern int nr_process, nr_thread;
458extern int shm_id;
459extern int groupid;
460extern int terse_output;
461extern int temp_stall_ts;
462extern unsigned long long mlock_size;
463extern unsigned long page_mask, page_size;
464extern int read_only;
465extern int eta_print;
466extern unsigned long done_secs;
467extern char *job_section;
468extern int fio_gtod_offload;
469extern int fio_gtod_cpu;
470
471extern struct thread_data *threads;
472
473static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
474{
475 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
476}
477
478#define BLOCKS_PER_MAP (8 * sizeof(int))
479#define TO_MAP_BLOCK(f, b) (b)
480#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
481#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
482
483#define MAX_JOBS (1024)
484
485#define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
486
487static inline void update_error_count(struct thread_data *td, int err)
488{
489 td->total_err_count++;
490 if (td->total_err_count == 1)
491 td->first_error = err;
492}
493
494static inline int should_fsync(struct thread_data *td)
495{
496 if (td->last_was_sync)
497 return 0;
498 if (td->o.odirect)
499 return 0;
500 if (td_write(td) || td_rw(td) || td->o.override_sync)
501 return 1;
502
503 return 0;
504}
505
506/*
507 * Time functions
508 */
509extern unsigned long long utime_since(struct timeval *, struct timeval *);
510extern unsigned long long utime_since_now(struct timeval *);
511extern unsigned long mtime_since(struct timeval *, struct timeval *);
512extern unsigned long mtime_since_now(struct timeval *);
513extern unsigned long time_since_now(struct timeval *);
514extern unsigned long mtime_since_genesis(void);
515extern void usec_spin(unsigned int);
516extern void usec_sleep(struct thread_data *, unsigned long);
517extern void fill_start_time(struct timeval *);
518extern void fio_gettime(struct timeval *, void *);
519extern void fio_gtod_init(void);
520extern void fio_gtod_update(void);
521extern void set_genesis_time(void);
522extern int ramp_time_over(struct thread_data *);
523extern int in_ramp_time(struct thread_data *);
524
525/*
526 * Init/option functions
527 */
528extern int __must_check parse_options(int, char **);
529extern int fio_options_parse(struct thread_data *, char **, int);
530extern void fio_keywords_init(void);
531extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
532extern void fio_fill_default_options(struct thread_data *);
533extern int fio_show_option_help(const char *);
534extern void fio_options_dup_and_init(struct option *);
535extern void options_mem_dupe(struct thread_data *);
536extern void options_mem_free(struct thread_data *);
537extern void td_fill_rand_seeds(struct thread_data *);
538#define FIO_GETOPT_JOB 0x89988998
539#define FIO_NR_OPTIONS 128
540
541/*
542 * ETA/status stuff
543 */
544extern void print_thread_status(void);
545extern void print_status_init(int);
546
547/*
548 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
549 * will never back again. It may cycle between running/verififying/fsyncing.
550 * Once the thread reaches TD_EXITED, it is just waiting for the core to
551 * reap it.
552 */
553enum {
554 TD_NOT_CREATED = 0,
555 TD_CREATED,
556 TD_INITIALIZED,
557 TD_RAMP,
558 TD_RUNNING,
559 TD_PRE_READING,
560 TD_VERIFYING,
561 TD_FSYNCING,
562 TD_EXITED,
563 TD_REAPED,
564};
565
566extern void td_set_runstate(struct thread_data *, int);
567
568/*
569 * Memory helpers
570 */
571extern int __must_check fio_pin_memory(void);
572extern void fio_unpin_memory(void);
573extern int __must_check allocate_io_mem(struct thread_data *);
574extern void free_io_mem(struct thread_data *);
575
576/*
577 * Reset stats after ramp time completes
578 */
579extern void reset_all_stats(struct thread_data *);
580
581/*
582 * blktrace support
583 */
584#ifdef FIO_HAVE_BLKTRACE
585extern int is_blktrace(const char *);
586extern int load_blktrace(struct thread_data *, const char *);
587#endif
588
589/*
590 * Mark unused variables passed to ops functions as unused, to silence gcc
591 */
592#define fio_unused __attribute((__unused__))
593#define fio_init __attribute__((constructor))
594#define fio_exit __attribute__((destructor))
595
596#define for_each_td(td, i) \
597 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
598#define for_each_file(td, f, i) \
599 if ((td)->files_index) \
600 for ((i) = 0, (f) = (td)->files[0]; \
601 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
602 (i)++)
603
604#define fio_assert(td, cond) do { \
605 if (!(cond)) { \
606 int *__foo = NULL; \
607 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
608 (td)->runstate = TD_EXITED; \
609 (td)->error = EFAULT; \
610 *__foo = 0; \
611 } \
612} while (0)
613
614static inline int fio_fill_issue_time(struct thread_data *td)
615{
616 if (td->o.read_iolog_file ||
617 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
618 return 1;
619
620 return 0;
621}
622
623/*
624 * Cheesy number->string conversion, complete with carry rounding error.
625 */
626static inline char *num2str(unsigned long num, int maxlen, int base, int pow2)
627{
628 char postfix[] = { ' ', 'K', 'M', 'G', 'P', 'E' };
629 unsigned int thousand;
630 char *buf;
631 int i;
632
633 if (pow2)
634 thousand = 1024;
635 else
636 thousand = 1000;
637
638 buf = malloc(128);
639
640 for (i = 0; base > 1; i++)
641 base /= thousand;
642
643 do {
644 int len, carry = 0;
645
646 len = sprintf(buf, "%'lu", num);
647 if (len <= maxlen) {
648 if (i >= 1) {
649 buf[len] = postfix[i];
650 buf[len + 1] = '\0';
651 }
652 return buf;
653 }
654
655 if ((num % thousand) >= (thousand / 2))
656 carry = 1;
657
658 num /= thousand;
659 num += carry;
660 i++;
661 } while (i <= 5);
662
663 return buf;
664}
665
666static inline int __should_check_rate(struct thread_data *td,
667 enum fio_ddir ddir)
668{
669 struct thread_options *o = &td->o;
670
671 /*
672 * If some rate setting was given, we need to check it
673 */
674 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
675 o->rate_iops_min[ddir])
676 return 1;
677
678 return 0;
679}
680
681static inline int should_check_rate(struct thread_data *td,
682 unsigned long *bytes_done)
683{
684 int ret = 0;
685
686 if (bytes_done[0])
687 ret |= __should_check_rate(td, 0);
688 if (bytes_done[1])
689 ret |= __should_check_rate(td, 1);
690
691 return ret;
692}
693
694static inline int is_power_of_2(unsigned int val)
695{
696 return (val != 0 && ((val & (val - 1)) == 0));
697}
698
699/*
700 * We currently only need to do locking if we have verifier threads
701 * accessing our internal structures too
702 */
703static inline void td_io_u_lock(struct thread_data *td)
704{
705 if (td->o.verify_async)
706 pthread_mutex_lock(&td->io_u_lock);
707}
708
709static inline void td_io_u_unlock(struct thread_data *td)
710{
711 if (td->o.verify_async)
712 pthread_mutex_unlock(&td->io_u_lock);
713}
714
715static inline void td_io_u_free_notify(struct thread_data *td)
716{
717 if (td->o.verify_async)
718 pthread_cond_signal(&td->free_cond);
719}
720
721#endif
fio - Flexible IO Tester