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>
51 #ifdef FIO_HAVE_LIBNUMA
52 #include <linux/mempolicy.h>
56 * "local" is pseudo-policy
58 #define MPOL_LOCAL MPOL_MAX
62 * What type of allocation to use for io buffers
65 MEM_MALLOC = 0, /* ordinary malloc */
66 MEM_SHM, /* use shared memory segments */
67 MEM_SHMHUGE, /* use shared memory segments with huge pages */
68 MEM_MMAP, /* use anonynomous mmap */
69 MEM_MMAPHUGE, /* memory mapped huge file */
73 * offset generator types
81 * What type of errors to continue on when continue_on_error is used
84 ERROR_TYPE_READ_BIT = 0,
85 ERROR_TYPE_WRITE_BIT = 1,
86 ERROR_TYPE_VERIFY_BIT = 2,
92 ERROR_TYPE_READ = 1 << ERROR_TYPE_READ_BIT,
93 ERROR_TYPE_WRITE = 1 << ERROR_TYPE_WRITE_BIT,
94 ERROR_TYPE_VERIFY = 1 << ERROR_TYPE_VERIFY_BIT,
95 ERROR_TYPE_ANY = 0xffff,
103 struct thread_options {
111 enum td_ddir td_ddir;
113 unsigned int kb_base;
114 unsigned int ddir_seq_nr;
116 unsigned int iodepth;
117 unsigned int iodepth_low;
118 unsigned int iodepth_batch;
119 unsigned int iodepth_batch_complete;
121 unsigned long long size;
122 unsigned int size_percent;
123 unsigned int fill_device;
124 unsigned long long file_size_low;
125 unsigned long long file_size_high;
126 unsigned long long start_offset;
128 unsigned int bs[DDIR_RWDIR_CNT];
129 unsigned int ba[DDIR_RWDIR_CNT];
130 unsigned int min_bs[DDIR_RWDIR_CNT];
131 unsigned int max_bs[DDIR_RWDIR_CNT];
132 struct bssplit *bssplit[DDIR_RWDIR_CNT];
133 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
135 int *ignore_error[ERROR_TYPE_CNT];
136 unsigned int ignore_error_nr[ERROR_TYPE_CNT];
137 unsigned int error_dump;
139 unsigned int nr_files;
140 unsigned int open_files;
141 enum file_lock_mode file_lock_mode;
142 unsigned int lockfile_batch;
144 unsigned int odirect;
145 unsigned int invalidate_cache;
146 unsigned int create_serialize;
147 unsigned int create_fsync;
148 unsigned int create_on_open;
149 unsigned int create_only;
150 unsigned int end_fsync;
151 unsigned int pre_read;
152 unsigned int sync_io;
154 unsigned int do_verify;
155 unsigned int verifysort;
156 unsigned int verify_interval;
157 unsigned int verify_offset;
158 char verify_pattern[MAX_PATTERN_SIZE];
159 unsigned int verify_pattern_bytes;
160 unsigned int verify_fatal;
161 unsigned int verify_dump;
162 unsigned int verify_async;
163 unsigned long long verify_backlog;
164 unsigned int verify_batch;
165 unsigned int use_thread;
167 unsigned int do_disk_util;
168 unsigned int override_sync;
169 unsigned int rand_repeatable;
170 unsigned int use_os_rand;
171 unsigned int write_lat_log;
172 unsigned int write_bw_log;
173 unsigned int write_iops_log;
174 unsigned int log_avg_msec;
175 unsigned int norandommap;
176 unsigned int softrandommap;
177 unsigned int bs_unaligned;
178 unsigned int fsync_on_close;
180 unsigned int hugepage_size;
181 unsigned int rw_min_bs;
182 unsigned int thinktime;
183 unsigned int thinktime_spin;
184 unsigned int thinktime_blocks;
185 unsigned int fsync_blocks;
186 unsigned int fdatasync_blocks;
187 unsigned int barrier_blocks;
188 unsigned long long start_delay;
189 unsigned long long timeout;
190 unsigned long long ramp_time;
191 unsigned int overwrite;
192 unsigned int bw_avg_time;
193 unsigned int iops_avg_time;
195 unsigned long long zone_range;
196 unsigned long long zone_size;
197 unsigned long long zone_skip;
198 enum fio_memtype mem_type;
199 unsigned int mem_align;
201 unsigned int stonewall;
202 unsigned int new_group;
203 unsigned int numjobs;
204 os_cpu_mask_t cpumask;
205 unsigned int cpumask_set;
206 os_cpu_mask_t verify_cpumask;
207 unsigned int verify_cpumask_set;
208 #ifdef FIO_HAVE_LIBNUMA
209 struct bitmask *numa_cpunodesmask;
210 unsigned int numa_cpumask_set;
211 unsigned short numa_mem_mode;
212 unsigned int numa_mem_prefer_node;
213 struct bitmask *numa_memnodesmask;
214 unsigned int numa_memmask_set;
217 unsigned int rwmixcycle;
218 unsigned int rwmix[2];
220 unsigned int file_service_type;
221 unsigned int group_reporting;
222 unsigned int fadvise_hint;
223 enum fio_fallocate_mode fallocate_mode;
224 unsigned int zero_buffers;
225 unsigned int refill_buffers;
226 unsigned int scramble_buffers;
227 unsigned int compress_percentage;
228 unsigned int compress_chunk;
229 unsigned int time_based;
230 unsigned int disable_lat;
231 unsigned int disable_clat;
232 unsigned int disable_slat;
233 unsigned int disable_bw;
234 unsigned int gtod_reduce;
235 unsigned int gtod_cpu;
236 unsigned int gtod_offload;
237 enum fio_cs clocksource;
238 unsigned int no_stall;
239 unsigned int trim_percentage;
240 unsigned int trim_batch;
241 unsigned int trim_zero;
242 unsigned long long trim_backlog;
243 unsigned int clat_percentiles;
244 unsigned int overwrite_plist;
245 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
247 char *read_iolog_file;
248 char *write_iolog_file;
252 char *replay_redirect;
255 * Pre-run and post-run shell
260 unsigned int rate[DDIR_RWDIR_CNT];
261 unsigned int ratemin[DDIR_RWDIR_CNT];
262 unsigned int ratecycle;
263 unsigned int rate_iops[DDIR_RWDIR_CNT];
264 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
269 * CPU "io" cycle burner
271 unsigned int cpuload;
272 unsigned int cpucycle;
277 enum error_type continue_on_error;
280 * Benchmark profile type
285 * blkio cgroup support
288 unsigned int cgroup_weight;
289 unsigned int cgroup_nodelete;
297 unsigned int flow_sleep;
299 unsigned long long offset_increment;
301 unsigned int sync_file_range;
305 * This describes a single thread/process executing a fio job.
308 struct thread_options o;
310 char verror[FIO_VERROR_SIZE];
314 struct thread_stat ts;
316 struct io_log *slat_log;
317 struct io_log *clat_log;
318 struct io_log *lat_log;
319 struct io_log *bw_log;
320 struct io_log *iops_log;
322 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
323 struct timeval bw_sample_time;
325 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
326 struct timeval iops_sample_time;
328 struct rusage ru_start;
329 struct rusage ru_end;
331 struct fio_file **files;
332 unsigned int files_size;
333 unsigned int files_index;
334 unsigned int nr_open_files;
335 unsigned int nr_done_files;
336 unsigned int nr_normal_files;
338 unsigned int next_file;
339 os_random_state_t next_file_state;
340 struct frand_state __next_file_state;
347 size_t orig_buffer_size;
348 volatile int terminate;
349 volatile int runstate;
351 unsigned int ioprio_set;
352 unsigned int last_was_sync;
353 enum fio_ddir last_ddir;
363 unsigned long rand_seeds[8];
366 os_random_state_t bsrange_state;
367 struct frand_state __bsrange_state;
370 os_random_state_t verify_state;
371 struct frand_state __verify_state;
374 os_random_state_t trim_state;
375 struct frand_state __trim_state;
378 struct frand_state buf_state;
380 unsigned int verify_batch;
381 unsigned int trim_batch;
386 * IO engine hooks, contains everything needed to submit an io_u
387 * to any of the available IO engines.
389 struct ioengine_ops *io_ops;
392 * Queue depth of io_u's that fio MIGHT do
394 unsigned int cur_depth;
397 * io_u's about to be committed
399 unsigned int io_u_queued;
402 * io_u's submitted but not completed yet
404 unsigned int io_u_in_flight;
407 * List of free and busy io_u's
409 struct flist_head io_u_freelist;
410 struct flist_head io_u_busylist;
411 struct flist_head io_u_requeues;
412 pthread_mutex_t io_u_lock;
413 pthread_cond_t free_cond;
416 * async verify offload
418 struct flist_head verify_list;
419 pthread_t *verify_threads;
420 unsigned int nr_verify_threads;
421 pthread_cond_t verify_cond;
422 int verify_thread_exit;
427 unsigned long long rate_bps[DDIR_RWDIR_CNT];
428 long rate_pending_usleep[DDIR_RWDIR_CNT];
429 unsigned long rate_bytes[DDIR_RWDIR_CNT];
430 unsigned long rate_blocks[DDIR_RWDIR_CNT];
431 struct timeval lastrate[DDIR_RWDIR_CNT];
433 unsigned long long total_io_size;
434 unsigned long long fill_device_size;
436 unsigned long io_issues[DDIR_RWDIR_CNT];
437 unsigned long long io_blocks[DDIR_RWDIR_CNT];
438 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
439 unsigned long long io_bytes[DDIR_RWDIR_CNT];
440 unsigned long long io_skip_bytes;
441 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
442 unsigned long long zone_bytes;
443 struct fio_mutex *mutex;
446 * State for random io, a bitmap of blocks done vs not done
449 os_random_state_t random_state;
450 struct frand_state __random_state;
453 struct timeval start; /* start of this loop */
454 struct timeval epoch; /* time job was started */
455 struct timeval last_issue;
456 struct timeval tv_cache;
457 unsigned int tv_cache_nr;
458 unsigned int tv_cache_mask;
459 unsigned int ramp_time_over;
462 * read/write mixed workload state
465 os_random_state_t rwmix_state;
466 struct frand_state __rwmix_state;
468 unsigned long rwmix_issues;
469 enum fio_ddir rwmix_ddir;
470 unsigned int ddir_seq_nr;
473 * IO history logs for verification. We use a tree for sorting,
474 * if we are overwriting. Otherwise just use a fifo.
476 struct rb_root io_hist_tree;
477 struct flist_head io_hist_list;
478 unsigned long io_hist_len;
483 struct flist_head io_log_list;
486 * For tracking/handling discards
488 struct flist_head trim_list;
489 unsigned long trim_entries;
492 * for fileservice, how often to switch to a new file
494 unsigned int file_service_nr;
495 unsigned int file_service_left;
496 struct fio_file *file_service_file;
498 unsigned int sync_file_range_nr;
501 * For generating file sizes
504 os_random_state_t file_size_state;
505 struct frand_state __file_size_state;
511 unsigned int total_err_count;
514 struct fio_flow *flow;
517 * Can be overloaded by profiles
519 struct prof_io_ops prof_io_ops;
524 * when should interactive ETA output be generated
532 #define __td_verror(td, err, msg, func) \
538 if (!(td)->first_error) \
539 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
543 #define td_clear_error(td) \
545 #define td_verror(td, err, func) \
546 __td_verror((td), (err), strerror((err)), (func))
547 #define td_vmsg(td, err, msg, func) \
548 __td_verror((td), (err), (msg), (func))
550 #define __fio_stringify_1(x) #x
551 #define __fio_stringify(x) __fio_stringify_1(x)
553 extern int exitall_on_terminate;
554 extern unsigned int thread_number;
555 extern unsigned int nr_process, nr_thread;
558 extern int output_format;
559 extern int temp_stall_ts;
560 extern unsigned long long mlock_size;
561 extern uintptr_t page_mask, page_size;
562 extern int read_only;
563 extern int eta_print;
564 extern unsigned long done_secs;
565 extern char *job_section;
566 extern int fio_gtod_offload;
567 extern int fio_gtod_cpu;
568 extern enum fio_cs fio_clock_source;
569 extern int warnings_fatal;
570 extern int terse_version;
571 extern int is_backend;
572 extern int nr_clients;
573 extern int log_syslog;
574 extern const char fio_version_string[];
575 extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
577 extern struct thread_data *threads;
579 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
581 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
584 #define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
585 #define TO_MAP_BLOCK(f, b) (b)
586 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
587 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
589 #define REAL_MAX_JOBS 2048
591 static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
594 return ERROR_TYPE_VERIFY_BIT;
595 if (ddir == DDIR_READ)
596 return ERROR_TYPE_READ_BIT;
597 return ERROR_TYPE_WRITE_BIT;
600 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
601 static inline int td_non_fatal_error(struct thread_data *td,
602 enum error_type_bit etype, int err)
605 if (!td->o.ignore_error[etype]) {
606 td->o.ignore_error[etype] = __NON_FATAL_ERR;
607 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
611 if (!(td->o.continue_on_error & (1 << etype)))
613 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
614 if (td->o.ignore_error[etype][i] == err)
619 static inline void update_error_count(struct thread_data *td, int err)
621 td->total_err_count++;
622 if (td->total_err_count == 1)
623 td->first_error = err;
626 static inline int should_fsync(struct thread_data *td)
628 if (td->last_was_sync)
632 if (td_write(td) || td_rw(td) || td->o.override_sync)
639 * Init/option functions
641 extern int __must_check parse_options(int, char **);
642 extern int parse_jobs_ini(char *, int, int);
643 extern int parse_cmd_line(int, char **);
644 extern int fio_backend(void);
645 extern void reset_fio_state(void);
646 extern void clear_io_state(struct thread_data *);
647 extern int fio_options_parse(struct thread_data *, char **, int);
648 extern void fio_keywords_init(void);
649 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
650 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
651 extern void fio_fill_default_options(struct thread_data *);
652 extern int fio_show_option_help(const char *);
653 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
654 extern void fio_options_dup_and_init(struct option *);
655 extern void fio_options_mem_dupe(struct thread_data *);
656 extern void options_mem_dupe(void *data, struct fio_option *options);
657 extern void td_fill_rand_seeds(struct thread_data *);
658 extern void add_job_opts(const char **);
659 extern char *num2str(unsigned long, int, int, int);
660 extern int ioengine_load(struct thread_data *);
662 #define FIO_GETOPT_JOB 0x89000000
663 #define FIO_GETOPT_IOENGINE 0x98000000
664 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
669 extern void print_thread_status(void);
670 extern void print_status_init(int);
673 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
674 * will never back again. It may cycle between running/verififying/fsyncing.
675 * Once the thread reaches TD_EXITED, it is just waiting for the core to
691 extern void td_set_runstate(struct thread_data *, int);
692 #define TERMINATE_ALL (-1)
693 extern void fio_terminate_threads(int);
698 extern int __must_check fio_pin_memory(void);
699 extern void fio_unpin_memory(void);
700 extern int __must_check allocate_io_mem(struct thread_data *);
701 extern void free_io_mem(struct thread_data *);
704 * Reset stats after ramp time completes
706 extern void reset_all_stats(struct thread_data *);
711 #ifdef FIO_HAVE_BLKTRACE
712 extern int is_blktrace(const char *);
713 extern int load_blktrace(struct thread_data *, const char *);
717 * Mark unused variables passed to ops functions as unused, to silence gcc
719 #define fio_unused __attribute((__unused__))
720 #define fio_init __attribute__((constructor))
721 #define fio_exit __attribute__((destructor))
723 #define for_each_td(td, i) \
724 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
725 #define for_each_file(td, f, i) \
726 if ((td)->files_index) \
727 for ((i) = 0, (f) = (td)->files[0]; \
728 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
731 #define fio_assert(td, cond) do { \
734 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
735 td_set_runstate((td), TD_EXITED); \
736 (td)->error = EFAULT; \
741 static inline int fio_fill_issue_time(struct thread_data *td)
743 if (td->o.read_iolog_file ||
744 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
750 static inline int __should_check_rate(struct thread_data *td,
753 struct thread_options *o = &td->o;
756 * If some rate setting was given, we need to check it
758 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
759 o->rate_iops_min[ddir])
765 static inline int should_check_rate(struct thread_data *td,
766 unsigned long *bytes_done)
770 if (bytes_done[DDIR_READ])
771 ret |= __should_check_rate(td, DDIR_READ);
772 if (bytes_done[DDIR_WRITE])
773 ret |= __should_check_rate(td, DDIR_WRITE);
774 if (bytes_done[DDIR_TRIM])
775 ret |= __should_check_rate(td, DDIR_TRIM);
780 static inline int is_power_of_2(unsigned int val)
782 return (val != 0 && ((val & (val - 1)) == 0));
786 * We currently only need to do locking if we have verifier threads
787 * accessing our internal structures too
789 static inline void td_io_u_lock(struct thread_data *td)
791 if (td->o.verify_async)
792 pthread_mutex_lock(&td->io_u_lock);
795 static inline void td_io_u_unlock(struct thread_data *td)
797 if (td->o.verify_async)
798 pthread_mutex_unlock(&td->io_u_lock);
801 static inline void td_io_u_free_notify(struct thread_data *td)
803 if (td->o.verify_async)
804 pthread_cond_signal(&td->free_cond);
807 extern const char *fio_get_arch_string(int);
808 extern const char *fio_get_os_string(int);
811 FIO_OUTPUT_TERSE = 0,