8 #include <sys/resource.h>
19 #include "compiler/compiler.h"
23 #include "arch/arch.h"
37 #include "lib/getopt.h"
47 #ifdef FIO_HAVE_SOLARISAIO
48 #include <sys/asynch.h>
52 * What type of allocation to use for io buffers
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
63 * offset generator types
71 * What type of errors to continue on when continue_on_error is used
75 ERROR_TYPE_READ = 1 << 0,
76 ERROR_TYPE_WRITE = 1 << 1,
77 ERROR_TYPE_VERIFY = 1 << 2,
78 ERROR_TYPE_ANY = 0xffff,
86 struct thread_options {
97 unsigned int ddir_seq_nr;
100 unsigned int iodepth_low;
101 unsigned int iodepth_batch;
102 unsigned int iodepth_batch_complete;
104 unsigned long long size;
105 unsigned int size_percent;
106 unsigned int fill_device;
107 unsigned long long file_size_low;
108 unsigned long long file_size_high;
109 unsigned long long start_offset;
113 unsigned int min_bs[2];
114 unsigned int max_bs[2];
115 struct bssplit *bssplit[2];
116 unsigned int bssplit_nr[2];
118 unsigned int nr_files;
119 unsigned int open_files;
120 enum file_lock_mode file_lock_mode;
121 unsigned int lockfile_batch;
123 unsigned int odirect;
124 unsigned int invalidate_cache;
125 unsigned int create_serialize;
126 unsigned int create_fsync;
127 unsigned int create_on_open;
128 unsigned int end_fsync;
129 unsigned int pre_read;
130 unsigned int sync_io;
132 unsigned int do_verify;
133 unsigned int verifysort;
134 unsigned int verify_interval;
135 unsigned int verify_offset;
136 char verify_pattern[MAX_PATTERN_SIZE];
137 unsigned int verify_pattern_bytes;
138 unsigned int verify_fatal;
139 unsigned int verify_dump;
140 unsigned int verify_async;
141 unsigned long long verify_backlog;
142 unsigned int verify_batch;
143 unsigned int use_thread;
145 unsigned int do_disk_util;
146 unsigned int override_sync;
147 unsigned int rand_repeatable;
148 unsigned int use_os_rand;
149 unsigned int write_lat_log;
150 unsigned int write_bw_log;
151 unsigned int write_iops_log;
152 unsigned int log_avg_msec;
153 unsigned int norandommap;
154 unsigned int softrandommap;
155 unsigned int bs_unaligned;
156 unsigned int fsync_on_close;
158 unsigned int hugepage_size;
159 unsigned int rw_min_bs;
160 unsigned int thinktime;
161 unsigned int thinktime_spin;
162 unsigned int thinktime_blocks;
163 unsigned int fsync_blocks;
164 unsigned int fdatasync_blocks;
165 unsigned int barrier_blocks;
166 unsigned long long start_delay;
167 unsigned long long timeout;
168 unsigned long long ramp_time;
169 unsigned int overwrite;
170 unsigned int bw_avg_time;
171 unsigned int iops_avg_time;
173 unsigned long long zone_range;
174 unsigned long long zone_size;
175 unsigned long long zone_skip;
176 enum fio_memtype mem_type;
177 unsigned int mem_align;
179 unsigned int stonewall;
180 unsigned int new_group;
181 unsigned int numjobs;
182 os_cpu_mask_t cpumask;
183 unsigned int cpumask_set;
184 os_cpu_mask_t verify_cpumask;
185 unsigned int verify_cpumask_set;
187 unsigned int rwmixcycle;
188 unsigned int rwmix[2];
190 unsigned int file_service_type;
191 unsigned int group_reporting;
192 unsigned int fadvise_hint;
193 enum fio_fallocate_mode fallocate_mode;
194 unsigned int zero_buffers;
195 unsigned int refill_buffers;
196 unsigned int scramble_buffers;
197 unsigned int compress_percentage;
198 unsigned int compress_chunk;
199 unsigned int time_based;
200 unsigned int disable_lat;
201 unsigned int disable_clat;
202 unsigned int disable_slat;
203 unsigned int disable_bw;
204 unsigned int gtod_reduce;
205 unsigned int gtod_cpu;
206 unsigned int gtod_offload;
207 enum fio_cs clocksource;
208 unsigned int no_stall;
209 unsigned int trim_percentage;
210 unsigned int trim_batch;
211 unsigned int trim_zero;
212 unsigned long long trim_backlog;
213 unsigned int clat_percentiles;
214 unsigned int overwrite_plist;
215 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
217 char *read_iolog_file;
218 char *write_iolog_file;
222 char *replay_redirect;
225 * Pre-run and post-run shell
230 unsigned int rate[2];
231 unsigned int ratemin[2];
232 unsigned int ratecycle;
233 unsigned int rate_iops[2];
234 unsigned int rate_iops_min[2];
239 * CPU "io" cycle burner
241 unsigned int cpuload;
242 unsigned int cpucycle;
247 enum error_type continue_on_error;
250 * Benchmark profile type
255 * blkio cgroup support
258 unsigned int cgroup_weight;
259 unsigned int cgroup_nodelete;
267 unsigned int flow_sleep;
269 unsigned int sync_file_range;
273 * This describes a single thread/process executing a fio job.
276 struct thread_options o;
278 char verror[FIO_VERROR_SIZE];
282 struct thread_stat ts;
284 struct io_log *slat_log;
285 struct io_log *clat_log;
286 struct io_log *lat_log;
287 struct io_log *bw_log;
288 struct io_log *iops_log;
290 uint64_t stat_io_bytes[2];
291 struct timeval bw_sample_time;
293 uint64_t stat_io_blocks[2];
294 struct timeval iops_sample_time;
296 struct rusage ru_start;
297 struct rusage ru_end;
299 struct fio_file **files;
300 unsigned int files_size;
301 unsigned int files_index;
302 unsigned int nr_open_files;
303 unsigned int nr_done_files;
304 unsigned int nr_normal_files;
306 unsigned int next_file;
307 os_random_state_t next_file_state;
308 struct frand_state __next_file_state;
314 size_t orig_buffer_size;
315 volatile int terminate;
316 volatile int runstate;
318 unsigned int ioprio_set;
319 unsigned int last_was_sync;
320 enum fio_ddir last_ddir;
330 unsigned long rand_seeds[8];
333 os_random_state_t bsrange_state;
334 struct frand_state __bsrange_state;
337 os_random_state_t verify_state;
338 struct frand_state __verify_state;
341 os_random_state_t trim_state;
342 struct frand_state __trim_state;
345 struct frand_state buf_state;
347 unsigned int verify_batch;
348 unsigned int trim_batch;
353 * IO engine hooks, contains everything needed to submit an io_u
354 * to any of the available IO engines.
356 struct ioengine_ops *io_ops;
359 * Queue depth of io_u's that fio MIGHT do
361 unsigned int cur_depth;
364 * io_u's about to be committed
366 unsigned int io_u_queued;
369 * io_u's submitted but not completed yet
371 unsigned int io_u_in_flight;
374 * List of free and busy io_u's
376 struct flist_head io_u_freelist;
377 struct flist_head io_u_busylist;
378 struct flist_head io_u_requeues;
379 pthread_mutex_t io_u_lock;
380 pthread_cond_t free_cond;
383 * async verify offload
385 struct flist_head verify_list;
386 pthread_t *verify_threads;
387 unsigned int nr_verify_threads;
388 pthread_cond_t verify_cond;
389 int verify_thread_exit;
394 unsigned long long rate_bps[2];
395 long rate_pending_usleep[2];
396 unsigned long rate_bytes[2];
397 unsigned long rate_blocks[2];
398 struct timeval lastrate[2];
400 unsigned long long total_io_size;
401 unsigned long long fill_device_size;
403 unsigned long io_issues[2];
404 unsigned long long io_blocks[2];
405 unsigned long long this_io_blocks[2];
406 unsigned long long io_bytes[2];
407 unsigned long long io_skip_bytes;
408 unsigned long long this_io_bytes[2];
409 unsigned long long zone_bytes;
410 struct fio_mutex *mutex;
413 * State for random io, a bitmap of blocks done vs not done
416 os_random_state_t random_state;
417 struct frand_state __random_state;
420 struct timeval start; /* start of this loop */
421 struct timeval epoch; /* time job was started */
422 struct timeval last_issue;
423 struct timeval tv_cache;
424 unsigned int tv_cache_nr;
425 unsigned int tv_cache_mask;
426 unsigned int ramp_time_over;
429 * read/write mixed workload state
432 os_random_state_t rwmix_state;
433 struct frand_state __rwmix_state;
435 unsigned long rwmix_issues;
436 enum fio_ddir rwmix_ddir;
437 unsigned int ddir_seq_nr;
440 * IO history logs for verification. We use a tree for sorting,
441 * if we are overwriting. Otherwise just use a fifo.
443 struct rb_root io_hist_tree;
444 struct flist_head io_hist_list;
445 unsigned long io_hist_len;
450 struct flist_head io_log_list;
453 * For tracking/handling discards
455 struct flist_head trim_list;
456 unsigned long trim_entries;
459 * for fileservice, how often to switch to a new file
461 unsigned int file_service_nr;
462 unsigned int file_service_left;
463 struct fio_file *file_service_file;
465 unsigned int sync_file_range_nr;
468 * For generating file sizes
471 os_random_state_t file_size_state;
472 struct frand_state __file_size_state;
478 unsigned int total_err_count;
481 struct fio_flow *flow;
484 * Can be overloaded by profiles
486 struct prof_io_ops prof_io_ops;
491 * when should interactive ETA output be generated
499 #define __td_verror(td, err, msg, func) \
505 if (!(td)->first_error) \
506 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
510 #define td_clear_error(td) \
512 #define td_verror(td, err, func) \
513 __td_verror((td), (err), strerror((err)), (func))
514 #define td_vmsg(td, err, msg, func) \
515 __td_verror((td), (err), (msg), (func))
517 #define __fio_stringify_1(x) #x
518 #define __fio_stringify(x) __fio_stringify_1(x)
520 extern int exitall_on_terminate;
521 extern unsigned int thread_number;
522 extern unsigned int nr_process, nr_thread;
525 extern int terse_output;
526 extern int temp_stall_ts;
527 extern unsigned long long mlock_size;
528 extern unsigned long page_mask, page_size;
529 extern int read_only;
530 extern int eta_print;
531 extern unsigned long done_secs;
532 extern char *job_section;
533 extern int fio_gtod_offload;
534 extern int fio_gtod_cpu;
535 extern enum fio_cs fio_clock_source;
536 extern int warnings_fatal;
537 extern int terse_version;
538 extern int is_backend;
539 extern int nr_clients;
540 extern int log_syslog;
541 extern const char fio_version_string[];
542 extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
544 extern struct thread_data *threads;
546 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
548 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
551 #define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
552 #define TO_MAP_BLOCK(f, b) (b)
553 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
554 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
556 #define REAL_MAX_JOBS 2048
558 #define td_non_fatal_error(e) ((e) == EIO || (e) == EILSEQ)
560 static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
563 return ERROR_TYPE_VERIFY;
564 if (ddir == DDIR_READ)
565 return ERROR_TYPE_READ;
566 return ERROR_TYPE_WRITE;
569 static inline void update_error_count(struct thread_data *td, int err)
571 td->total_err_count++;
572 if (td->total_err_count == 1)
573 td->first_error = err;
576 static inline int should_fsync(struct thread_data *td)
578 if (td->last_was_sync)
582 if (td_write(td) || td_rw(td) || td->o.override_sync)
589 * Init/option functions
591 extern int __must_check parse_options(int, char **);
592 extern int parse_jobs_ini(char *, int, int);
593 extern int parse_cmd_line(int, char **);
594 extern int fio_backend(void);
595 extern void reset_fio_state(void);
596 extern void clear_io_state(struct thread_data *);
597 extern int fio_options_parse(struct thread_data *, char **, int);
598 extern void fio_keywords_init(void);
599 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
600 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
601 extern void fio_fill_default_options(struct thread_data *);
602 extern int fio_show_option_help(const char *);
603 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
604 extern void fio_options_dup_and_init(struct option *);
605 extern void fio_options_mem_dupe(struct thread_data *);
606 extern void options_mem_dupe(void *data, struct fio_option *options);
607 extern void td_fill_rand_seeds(struct thread_data *);
608 extern void add_job_opts(const char **);
609 extern char *num2str(unsigned long, int, int, int);
610 extern int ioengine_load(struct thread_data *);
612 #define FIO_GETOPT_JOB 0x89000000
613 #define FIO_GETOPT_IOENGINE 0x98000000
614 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
619 extern void print_thread_status(void);
620 extern void print_status_init(int);
623 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
624 * will never back again. It may cycle between running/verififying/fsyncing.
625 * Once the thread reaches TD_EXITED, it is just waiting for the core to
641 extern void td_set_runstate(struct thread_data *, int);
642 #define TERMINATE_ALL (-1)
643 extern void fio_terminate_threads(int);
648 extern int __must_check fio_pin_memory(void);
649 extern void fio_unpin_memory(void);
650 extern int __must_check allocate_io_mem(struct thread_data *);
651 extern void free_io_mem(struct thread_data *);
654 * Reset stats after ramp time completes
656 extern void reset_all_stats(struct thread_data *);
661 #ifdef FIO_HAVE_BLKTRACE
662 extern int is_blktrace(const char *);
663 extern int load_blktrace(struct thread_data *, const char *);
667 * Mark unused variables passed to ops functions as unused, to silence gcc
669 #define fio_unused __attribute((__unused__))
670 #define fio_init __attribute__((constructor))
671 #define fio_exit __attribute__((destructor))
673 #define for_each_td(td, i) \
674 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
675 #define for_each_file(td, f, i) \
676 if ((td)->files_index) \
677 for ((i) = 0, (f) = (td)->files[0]; \
678 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
681 #define fio_assert(td, cond) do { \
684 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
685 td_set_runstate((td), TD_EXITED); \
686 (td)->error = EFAULT; \
691 static inline int fio_fill_issue_time(struct thread_data *td)
693 if (td->o.read_iolog_file ||
694 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
700 static inline int __should_check_rate(struct thread_data *td,
703 struct thread_options *o = &td->o;
706 * If some rate setting was given, we need to check it
708 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
709 o->rate_iops_min[ddir])
715 static inline int should_check_rate(struct thread_data *td,
716 unsigned long *bytes_done)
721 ret |= __should_check_rate(td, 0);
723 ret |= __should_check_rate(td, 1);
728 static inline int is_power_of_2(unsigned int val)
730 return (val != 0 && ((val & (val - 1)) == 0));
734 * We currently only need to do locking if we have verifier threads
735 * accessing our internal structures too
737 static inline void td_io_u_lock(struct thread_data *td)
739 if (td->o.verify_async)
740 pthread_mutex_lock(&td->io_u_lock);
743 static inline void td_io_u_unlock(struct thread_data *td)
745 if (td->o.verify_async)
746 pthread_mutex_unlock(&td->io_u_lock);
749 static inline void td_io_u_free_notify(struct thread_data *td)
751 if (td->o.verify_async)
752 pthread_cond_signal(&td->free_cond);
755 extern const char *fio_get_arch_string(int);
756 extern const char *fio_get_os_string(int);