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 random_distribution;
184 unsigned int random_generator;
186 unsigned int hugepage_size;
187 unsigned int rw_min_bs;
188 unsigned int thinktime;
189 unsigned int thinktime_spin;
190 unsigned int thinktime_blocks;
191 unsigned int fsync_blocks;
192 unsigned int fdatasync_blocks;
193 unsigned int barrier_blocks;
194 unsigned long long start_delay;
195 unsigned long long timeout;
196 unsigned long long ramp_time;
197 unsigned int overwrite;
198 unsigned int bw_avg_time;
199 unsigned int iops_avg_time;
201 unsigned long long zone_range;
202 unsigned long long zone_size;
203 unsigned long long zone_skip;
204 enum fio_memtype mem_type;
205 unsigned int mem_align;
207 unsigned int max_latency;
209 unsigned int stonewall;
210 unsigned int new_group;
211 unsigned int numjobs;
212 os_cpu_mask_t cpumask;
213 unsigned int cpumask_set;
214 os_cpu_mask_t verify_cpumask;
215 unsigned int verify_cpumask_set;
216 #ifdef FIO_HAVE_LIBNUMA
217 struct bitmask *numa_cpunodesmask;
218 unsigned int numa_cpumask_set;
219 unsigned short numa_mem_mode;
220 unsigned int numa_mem_prefer_node;
221 struct bitmask *numa_memnodesmask;
222 unsigned int numa_memmask_set;
225 unsigned int rwmixcycle;
226 unsigned int rwmix[2];
228 unsigned int file_service_type;
229 unsigned int group_reporting;
230 unsigned int fadvise_hint;
231 enum fio_fallocate_mode fallocate_mode;
232 unsigned int zero_buffers;
233 unsigned int refill_buffers;
234 unsigned int scramble_buffers;
235 unsigned int compress_percentage;
236 unsigned int compress_chunk;
237 unsigned int time_based;
238 unsigned int disable_lat;
239 unsigned int disable_clat;
240 unsigned int disable_slat;
241 unsigned int disable_bw;
242 unsigned int gtod_reduce;
243 unsigned int gtod_cpu;
244 unsigned int gtod_offload;
245 enum fio_cs clocksource;
246 unsigned int no_stall;
247 unsigned int trim_percentage;
248 unsigned int trim_batch;
249 unsigned int trim_zero;
250 unsigned long long trim_backlog;
251 unsigned int clat_percentiles;
252 unsigned int overwrite_plist;
253 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
255 char *read_iolog_file;
256 char *write_iolog_file;
260 char *replay_redirect;
263 * Pre-run and post-run shell
268 unsigned int rate[DDIR_RWDIR_CNT];
269 unsigned int ratemin[DDIR_RWDIR_CNT];
270 unsigned int ratecycle;
271 unsigned int rate_iops[DDIR_RWDIR_CNT];
272 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
277 * CPU "io" cycle burner
279 unsigned int cpuload;
280 unsigned int cpucycle;
285 enum error_type continue_on_error;
288 * Benchmark profile type
293 * blkio cgroup support
296 unsigned int cgroup_weight;
297 unsigned int cgroup_nodelete;
305 unsigned int flow_sleep;
307 unsigned long long offset_increment;
309 unsigned int sync_file_range;
313 TD_F_VER_BACKLOG = 1,
314 TD_F_TRIM_BACKLOG = 2,
316 TD_F_REFILL_BUFFERS = 8,
317 TD_F_SCRAMBLE_BUFFERS = 16,
319 TD_F_PROFILE_OPS = 64,
328 FIO_RAND_FILE_SIZE_OFF,
335 * This describes a single thread/process executing a fio job.
338 struct thread_options o;
341 char verror[FIO_VERROR_SIZE];
345 struct thread_stat ts;
347 struct io_log *slat_log;
348 struct io_log *clat_log;
349 struct io_log *lat_log;
350 struct io_log *bw_log;
351 struct io_log *iops_log;
353 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
354 struct timeval bw_sample_time;
356 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
357 struct timeval iops_sample_time;
359 struct rusage ru_start;
360 struct rusage ru_end;
362 struct fio_file **files;
363 unsigned int files_size;
364 unsigned int files_index;
365 unsigned int nr_open_files;
366 unsigned int nr_done_files;
367 unsigned int nr_normal_files;
369 unsigned int next_file;
370 os_random_state_t next_file_state;
371 struct frand_state __next_file_state;
378 size_t orig_buffer_size;
379 volatile int terminate;
380 volatile int runstate;
382 unsigned int ioprio_set;
383 unsigned int last_was_sync;
384 enum fio_ddir last_ddir;
394 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
397 os_random_state_t bsrange_state;
398 struct frand_state __bsrange_state;
401 os_random_state_t verify_state;
402 struct frand_state __verify_state;
405 os_random_state_t trim_state;
406 struct frand_state __trim_state;
409 struct frand_state buf_state;
411 unsigned int verify_batch;
412 unsigned int trim_batch;
417 * IO engine hooks, contains everything needed to submit an io_u
418 * to any of the available IO engines.
420 struct ioengine_ops *io_ops;
423 * Queue depth of io_u's that fio MIGHT do
425 unsigned int cur_depth;
428 * io_u's about to be committed
430 unsigned int io_u_queued;
433 * io_u's submitted but not completed yet
435 unsigned int io_u_in_flight;
438 * List of free and busy io_u's
440 struct flist_head io_u_freelist;
441 struct flist_head io_u_busylist;
442 struct flist_head io_u_requeues;
443 pthread_mutex_t io_u_lock;
444 pthread_cond_t free_cond;
447 * async verify offload
449 struct flist_head verify_list;
450 pthread_t *verify_threads;
451 unsigned int nr_verify_threads;
452 pthread_cond_t verify_cond;
453 int verify_thread_exit;
458 unsigned long long rate_bps[DDIR_RWDIR_CNT];
459 long rate_pending_usleep[DDIR_RWDIR_CNT];
460 unsigned long rate_bytes[DDIR_RWDIR_CNT];
461 unsigned long rate_blocks[DDIR_RWDIR_CNT];
462 struct timeval lastrate[DDIR_RWDIR_CNT];
464 unsigned long long total_io_size;
465 unsigned long long fill_device_size;
467 unsigned long io_issues[DDIR_RWDIR_CNT];
468 unsigned long long io_blocks[DDIR_RWDIR_CNT];
469 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
470 unsigned long long io_bytes[DDIR_RWDIR_CNT];
471 unsigned long long io_skip_bytes;
472 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
473 unsigned long long zone_bytes;
474 struct fio_mutex *mutex;
477 * State for random io, a bitmap of blocks done vs not done
480 os_random_state_t random_state;
481 struct frand_state __random_state;
484 struct timeval start; /* start of this loop */
485 struct timeval epoch; /* time job was started */
486 struct timeval last_issue;
487 struct timeval tv_cache;
488 unsigned int tv_cache_nr;
489 unsigned int tv_cache_mask;
490 unsigned int ramp_time_over;
493 * read/write mixed workload state
496 os_random_state_t rwmix_state;
497 struct frand_state __rwmix_state;
499 unsigned long rwmix_issues;
500 enum fio_ddir rwmix_ddir;
501 unsigned int ddir_seq_nr;
504 * IO history logs for verification. We use a tree for sorting,
505 * if we are overwriting. Otherwise just use a fifo.
507 struct rb_root io_hist_tree;
508 struct flist_head io_hist_list;
509 unsigned long io_hist_len;
514 struct flist_head io_log_list;
517 * For tracking/handling discards
519 struct flist_head trim_list;
520 unsigned long trim_entries;
523 * for fileservice, how often to switch to a new file
525 unsigned int file_service_nr;
526 unsigned int file_service_left;
527 struct fio_file *file_service_file;
529 unsigned int sync_file_range_nr;
532 * For generating file sizes
535 os_random_state_t file_size_state;
536 struct frand_state __file_size_state;
542 unsigned int total_err_count;
545 struct fio_flow *flow;
548 * Can be overloaded by profiles
550 struct prof_io_ops prof_io_ops;
555 * when should interactive ETA output be generated
563 #define __td_verror(td, err, msg, func) \
569 if (!(td)->first_error) \
570 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
574 #define td_clear_error(td) \
576 #define td_verror(td, err, func) \
577 __td_verror((td), (err), strerror((err)), (func))
578 #define td_vmsg(td, err, msg, func) \
579 __td_verror((td), (err), (msg), (func))
581 #define __fio_stringify_1(x) #x
582 #define __fio_stringify(x) __fio_stringify_1(x)
584 extern int exitall_on_terminate;
585 extern unsigned int thread_number;
586 extern unsigned int stat_number;
587 extern unsigned int nr_process, nr_thread;
590 extern int output_format;
591 extern int temp_stall_ts;
592 extern unsigned long long mlock_size;
593 extern uintptr_t page_mask, page_size;
594 extern int read_only;
595 extern int eta_print;
596 extern unsigned long done_secs;
597 extern char *job_section;
598 extern int fio_gtod_offload;
599 extern int fio_gtod_cpu;
600 extern enum fio_cs fio_clock_source;
601 extern int fio_clock_source_set;
602 extern int warnings_fatal;
603 extern int terse_version;
604 extern int is_backend;
605 extern int nr_clients;
606 extern int log_syslog;
607 extern const char fio_version_string[];
608 extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
610 extern struct thread_data *threads;
612 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
614 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
617 #define REAL_MAX_JOBS 2048
619 static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
622 return ERROR_TYPE_VERIFY_BIT;
623 if (ddir == DDIR_READ)
624 return ERROR_TYPE_READ_BIT;
625 return ERROR_TYPE_WRITE_BIT;
628 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
629 static inline int td_non_fatal_error(struct thread_data *td,
630 enum error_type_bit etype, int err)
633 if (!td->o.ignore_error[etype]) {
634 td->o.ignore_error[etype] = __NON_FATAL_ERR;
635 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
639 if (!(td->o.continue_on_error & (1 << etype)))
641 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
642 if (td->o.ignore_error[etype][i] == err)
647 static inline void update_error_count(struct thread_data *td, int err)
649 td->total_err_count++;
650 if (td->total_err_count == 1)
651 td->first_error = err;
654 static inline int should_fsync(struct thread_data *td)
656 if (td->last_was_sync)
660 if (td_write(td) || td_rw(td) || td->o.override_sync)
667 * Init/option functions
669 extern int __must_check parse_options(int, char **);
670 extern int parse_jobs_ini(char *, int, int);
671 extern int parse_cmd_line(int, char **);
672 extern int fio_backend(void);
673 extern void reset_fio_state(void);
674 extern void clear_io_state(struct thread_data *);
675 extern int fio_options_parse(struct thread_data *, char **, int);
676 extern void fio_keywords_init(void);
677 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
678 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
679 extern void fio_fill_default_options(struct thread_data *);
680 extern int fio_show_option_help(const char *);
681 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
682 extern void fio_options_dup_and_init(struct option *);
683 extern void fio_options_mem_dupe(struct thread_data *);
684 extern void options_mem_dupe(void *data, struct fio_option *options);
685 extern void td_fill_rand_seeds(struct thread_data *);
686 extern void add_job_opts(const char **);
687 extern char *num2str(unsigned long, int, int, int);
688 extern int ioengine_load(struct thread_data *);
690 #define FIO_GETOPT_JOB 0x89000000
691 #define FIO_GETOPT_IOENGINE 0x98000000
692 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
697 extern void print_thread_status(void);
698 extern void print_status_init(int);
701 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
702 * will never back again. It may cycle between running/verififying/fsyncing.
703 * Once the thread reaches TD_EXITED, it is just waiting for the core to
720 extern void td_set_runstate(struct thread_data *, int);
721 #define TERMINATE_ALL (-1)
722 extern void fio_terminate_threads(int);
727 extern int __must_check fio_pin_memory(void);
728 extern void fio_unpin_memory(void);
729 extern int __must_check allocate_io_mem(struct thread_data *);
730 extern void free_io_mem(struct thread_data *);
733 * Reset stats after ramp time completes
735 extern void reset_all_stats(struct thread_data *);
740 #ifdef FIO_HAVE_BLKTRACE
741 extern int is_blktrace(const char *);
742 extern int load_blktrace(struct thread_data *, const char *);
746 * Mark unused variables passed to ops functions as unused, to silence gcc
748 #define fio_unused __attribute((__unused__))
749 #define fio_init __attribute__((constructor))
750 #define fio_exit __attribute__((destructor))
752 #define for_each_td(td, i) \
753 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
754 #define for_each_file(td, f, i) \
755 if ((td)->files_index) \
756 for ((i) = 0, (f) = (td)->files[0]; \
757 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
760 #define fio_assert(td, cond) do { \
763 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
764 td_set_runstate((td), TD_EXITED); \
765 (td)->error = EFAULT; \
770 static inline int fio_fill_issue_time(struct thread_data *td)
772 if (td->o.read_iolog_file ||
773 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
779 static inline int __should_check_rate(struct thread_data *td,
782 struct thread_options *o = &td->o;
785 * If some rate setting was given, we need to check it
787 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
788 o->rate_iops_min[ddir])
794 static inline int should_check_rate(struct thread_data *td,
795 unsigned long *bytes_done)
799 if (bytes_done[DDIR_READ])
800 ret |= __should_check_rate(td, DDIR_READ);
801 if (bytes_done[DDIR_WRITE])
802 ret |= __should_check_rate(td, DDIR_WRITE);
803 if (bytes_done[DDIR_TRIM])
804 ret |= __should_check_rate(td, DDIR_TRIM);
809 static inline int is_power_of_2(unsigned int val)
811 return (val != 0 && ((val & (val - 1)) == 0));
815 * We currently only need to do locking if we have verifier threads
816 * accessing our internal structures too
818 static inline void td_io_u_lock(struct thread_data *td)
820 if (td->o.verify_async)
821 pthread_mutex_lock(&td->io_u_lock);
824 static inline void td_io_u_unlock(struct thread_data *td)
826 if (td->o.verify_async)
827 pthread_mutex_unlock(&td->io_u_lock);
830 static inline void td_io_u_free_notify(struct thread_data *td)
832 if (td->o.verify_async)
833 pthread_cond_signal(&td->free_cond);
836 extern const char *fio_get_arch_string(int);
837 extern const char *fio_get_os_string(int);
840 FIO_OUTPUT_TERSE = 0,
846 FIO_RAND_DIST_RANDOM = 0,
848 FIO_RAND_DIST_PARETO,
852 FIO_RAND_GEN_TAUSWORTHE = 0,