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