8 #include <sys/resource.h>
19 #include "compiler/compiler.h"
23 #include "arch/arch.h"
37 #include "lib/getopt.h"
43 #ifdef CONFIG_SOLARISAIO
44 #include <sys/asynch.h>
48 #include <linux/mempolicy.h>
52 * "local" is pseudo-policy
54 #define MPOL_LOCAL MPOL_MAX
58 * What type of allocation to use for io buffers
61 MEM_MALLOC = 0, /* ordinary malloc */
62 MEM_SHM, /* use shared memory segments */
63 MEM_SHMHUGE, /* use shared memory segments with huge pages */
64 MEM_MMAP, /* use anonynomous mmap */
65 MEM_MMAPHUGE, /* memory mapped huge file */
69 * offset generator types
77 * What type of errors to continue on when continue_on_error is used
80 ERROR_TYPE_READ_BIT = 0,
81 ERROR_TYPE_WRITE_BIT = 1,
82 ERROR_TYPE_VERIFY_BIT = 2,
88 ERROR_TYPE_READ = 1 << ERROR_TYPE_READ_BIT,
89 ERROR_TYPE_WRITE = 1 << ERROR_TYPE_WRITE_BIT,
90 ERROR_TYPE_VERIFY = 1 << ERROR_TYPE_VERIFY_BIT,
91 ERROR_TYPE_ANY = 0xffff,
99 struct thread_options {
105 char *filename_format;
108 enum td_ddir td_ddir;
110 unsigned int kb_base;
111 unsigned int unit_base;
112 unsigned int ddir_seq_nr;
114 unsigned int iodepth;
115 unsigned int iodepth_low;
116 unsigned int iodepth_batch;
117 unsigned int iodepth_batch_complete;
119 unsigned long long size;
120 unsigned int size_percent;
121 unsigned int fill_device;
122 unsigned long long file_size_low;
123 unsigned long long file_size_high;
124 unsigned long long start_offset;
126 unsigned int bs[DDIR_RWDIR_CNT];
127 unsigned int ba[DDIR_RWDIR_CNT];
128 unsigned int min_bs[DDIR_RWDIR_CNT];
129 unsigned int max_bs[DDIR_RWDIR_CNT];
130 struct bssplit *bssplit[DDIR_RWDIR_CNT];
131 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
133 int *ignore_error[ERROR_TYPE_CNT];
134 unsigned int ignore_error_nr[ERROR_TYPE_CNT];
135 unsigned int error_dump;
137 unsigned int nr_files;
138 unsigned int open_files;
139 enum file_lock_mode file_lock_mode;
141 unsigned int odirect;
142 unsigned int invalidate_cache;
143 unsigned int create_serialize;
144 unsigned int create_fsync;
145 unsigned int create_on_open;
146 unsigned int create_only;
147 unsigned int end_fsync;
148 unsigned int pre_read;
149 unsigned int sync_io;
151 unsigned int do_verify;
152 unsigned int verifysort;
153 unsigned int verifysort_nr;
154 unsigned int verify_interval;
155 unsigned int verify_offset;
156 char verify_pattern[MAX_PATTERN_SIZE];
157 unsigned int verify_pattern_bytes;
158 unsigned int verify_fatal;
159 unsigned int verify_dump;
160 unsigned int verify_async;
161 unsigned long long verify_backlog;
162 unsigned int verify_batch;
163 unsigned int experimental_verify;
164 unsigned int use_thread;
166 unsigned int do_disk_util;
167 unsigned int override_sync;
168 unsigned int rand_repeatable;
169 unsigned int use_os_rand;
170 unsigned int write_lat_log;
171 unsigned int write_bw_log;
172 unsigned int write_iops_log;
173 unsigned int log_avg_msec;
174 unsigned int norandommap;
175 unsigned int softrandommap;
176 unsigned int bs_unaligned;
177 unsigned int fsync_on_close;
179 unsigned int random_distribution;
183 unsigned int random_generator;
185 unsigned int hugepage_size;
186 unsigned int rw_min_bs;
187 unsigned int thinktime;
188 unsigned int thinktime_spin;
189 unsigned int thinktime_blocks;
190 unsigned int fsync_blocks;
191 unsigned int fdatasync_blocks;
192 unsigned int barrier_blocks;
193 unsigned long long start_delay;
194 unsigned long long timeout;
195 unsigned long long ramp_time;
196 unsigned int overwrite;
197 unsigned int bw_avg_time;
198 unsigned int iops_avg_time;
200 unsigned long long zone_range;
201 unsigned long long zone_size;
202 unsigned long long zone_skip;
203 enum fio_memtype mem_type;
204 unsigned int mem_align;
206 unsigned int max_latency;
208 unsigned int stonewall;
209 unsigned int new_group;
210 unsigned int numjobs;
211 os_cpu_mask_t cpumask;
212 unsigned int cpumask_set;
213 os_cpu_mask_t verify_cpumask;
214 unsigned int verify_cpumask_set;
215 #ifdef CONFIG_LIBNUMA
216 struct bitmask *numa_cpunodesmask;
217 unsigned int numa_cpumask_set;
218 unsigned short numa_mem_mode;
219 unsigned int numa_mem_prefer_node;
220 struct bitmask *numa_memnodesmask;
221 unsigned int numa_memmask_set;
224 unsigned int rwmixcycle;
225 unsigned int rwmix[2];
227 unsigned int file_service_type;
228 unsigned int group_reporting;
229 unsigned int fadvise_hint;
230 enum fio_fallocate_mode fallocate_mode;
231 unsigned int zero_buffers;
232 unsigned int refill_buffers;
233 unsigned int scramble_buffers;
234 unsigned int compress_percentage;
235 unsigned int compress_chunk;
236 unsigned int time_based;
237 unsigned int disable_lat;
238 unsigned int disable_clat;
239 unsigned int disable_slat;
240 unsigned int disable_bw;
241 unsigned int unified_rw_rep;
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 percentile_precision; /* digits after decimal for percentiles */
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 volatile int update_rusage;
360 struct fio_mutex *rusage_sem;
361 struct rusage ru_start;
362 struct rusage ru_end;
364 struct fio_file **files;
365 unsigned char *file_locks;
366 unsigned int files_size;
367 unsigned int files_index;
368 unsigned int nr_open_files;
369 unsigned int nr_done_files;
370 unsigned int nr_normal_files;
372 unsigned int next_file;
373 os_random_state_t next_file_state;
374 struct frand_state __next_file_state;
381 size_t orig_buffer_size;
382 volatile int terminate;
383 volatile int runstate;
385 unsigned int ioprio_set;
386 unsigned int last_was_sync;
387 enum fio_ddir last_ddir;
397 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
400 os_random_state_t bsrange_state;
401 struct frand_state __bsrange_state;
404 os_random_state_t verify_state;
405 struct frand_state __verify_state;
408 os_random_state_t trim_state;
409 struct frand_state __trim_state;
412 struct frand_state buf_state;
414 unsigned int verify_batch;
415 unsigned int trim_batch;
420 * IO engine hooks, contains everything needed to submit an io_u
421 * to any of the available IO engines.
423 struct ioengine_ops *io_ops;
426 * Queue depth of io_u's that fio MIGHT do
428 unsigned int cur_depth;
431 * io_u's about to be committed
433 unsigned int io_u_queued;
436 * io_u's submitted but not completed yet
438 unsigned int io_u_in_flight;
441 * List of free and busy io_u's
443 struct flist_head io_u_freelist;
444 struct flist_head io_u_busylist;
445 struct flist_head io_u_requeues;
446 pthread_mutex_t io_u_lock;
447 pthread_cond_t free_cond;
450 * async verify offload
452 struct flist_head verify_list;
453 pthread_t *verify_threads;
454 unsigned int nr_verify_threads;
455 pthread_cond_t verify_cond;
456 int verify_thread_exit;
461 uint64_t rate_bps[DDIR_RWDIR_CNT];
462 long rate_pending_usleep[DDIR_RWDIR_CNT];
463 unsigned long rate_bytes[DDIR_RWDIR_CNT];
464 unsigned long rate_blocks[DDIR_RWDIR_CNT];
465 struct timeval lastrate[DDIR_RWDIR_CNT];
467 uint64_t total_io_size;
468 uint64_t fill_device_size;
470 unsigned long io_issues[DDIR_RWDIR_CNT];
471 uint64_t io_blocks[DDIR_RWDIR_CNT];
472 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
473 uint64_t io_bytes[DDIR_RWDIR_CNT];
474 uint64_t io_skip_bytes;
475 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
477 struct fio_mutex *mutex;
480 * State for random io, a bitmap of blocks done vs not done
483 os_random_state_t random_state;
484 struct frand_state __random_state;
487 struct timeval start; /* start of this loop */
488 struct timeval epoch; /* time job was started */
489 struct timeval last_issue;
490 struct timeval tv_cache;
491 unsigned int tv_cache_nr;
492 unsigned int tv_cache_mask;
493 unsigned int ramp_time_over;
496 * read/write mixed workload state
499 os_random_state_t rwmix_state;
500 struct frand_state __rwmix_state;
502 unsigned long rwmix_issues;
503 enum fio_ddir rwmix_ddir;
504 unsigned int ddir_seq_nr;
507 * IO history logs for verification. We use a tree for sorting,
508 * if we are overwriting. Otherwise just use a fifo.
510 struct rb_root io_hist_tree;
511 struct flist_head io_hist_list;
512 unsigned long io_hist_len;
517 struct flist_head io_log_list;
520 * For tracking/handling discards
522 struct flist_head trim_list;
523 unsigned long trim_entries;
525 struct flist_head next_rand_list;
528 * for fileservice, how often to switch to a new file
530 unsigned int file_service_nr;
531 unsigned int file_service_left;
532 struct fio_file *file_service_file;
534 unsigned int sync_file_range_nr;
537 * For generating file sizes
540 os_random_state_t file_size_state;
541 struct frand_state __file_size_state;
547 unsigned int total_err_count;
550 struct fio_flow *flow;
553 * Can be overloaded by profiles
555 struct prof_io_ops prof_io_ops;
560 * when should interactive ETA output be generated
568 #define __td_verror(td, err, msg, func) \
574 if (!(td)->first_error) \
575 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
579 #define td_clear_error(td) \
581 #define td_verror(td, err, func) \
582 __td_verror((td), (err), strerror((err)), (func))
583 #define td_vmsg(td, err, msg, func) \
584 __td_verror((td), (err), (msg), (func))
586 #define __fio_stringify_1(x) #x
587 #define __fio_stringify(x) __fio_stringify_1(x)
589 extern int exitall_on_terminate;
590 extern unsigned int thread_number;
591 extern unsigned int stat_number;
592 extern unsigned int nr_process, nr_thread;
595 extern int output_format;
596 extern int temp_stall_ts;
597 extern unsigned long long mlock_size;
598 extern uintptr_t page_mask, page_size;
599 extern int read_only;
600 extern int eta_print;
601 extern int eta_new_line;
602 extern unsigned long done_secs;
603 extern char *job_section;
604 extern int fio_gtod_offload;
605 extern int fio_gtod_cpu;
606 extern enum fio_cs fio_clock_source;
607 extern int fio_clock_source_set;
608 extern int warnings_fatal;
609 extern int terse_version;
610 extern int is_backend;
611 extern int nr_clients;
612 extern int log_syslog;
613 extern const char fio_version_string[];
615 extern struct thread_data *threads;
617 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
619 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
622 #define REAL_MAX_JOBS 2048
624 static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
627 return ERROR_TYPE_VERIFY_BIT;
628 if (ddir == DDIR_READ)
629 return ERROR_TYPE_READ_BIT;
630 return ERROR_TYPE_WRITE_BIT;
633 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
634 static inline int td_non_fatal_error(struct thread_data *td,
635 enum error_type_bit etype, int err)
638 if (!td->o.ignore_error[etype]) {
639 td->o.ignore_error[etype] = __NON_FATAL_ERR;
640 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
644 if (!(td->o.continue_on_error & (1 << etype)))
646 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
647 if (td->o.ignore_error[etype][i] == err)
652 static inline void update_error_count(struct thread_data *td, int err)
654 td->total_err_count++;
655 if (td->total_err_count == 1)
656 td->first_error = err;
659 static inline int should_fsync(struct thread_data *td)
661 if (td->last_was_sync)
663 if (td_write(td) || td_rw(td) || td->o.override_sync)
670 * Init/option functions
672 extern int __must_check parse_options(int, char **);
673 extern int parse_jobs_ini(char *, int, int);
674 extern int parse_cmd_line(int, char **);
675 extern int fio_backend(void);
676 extern void reset_fio_state(void);
677 extern void clear_io_state(struct thread_data *);
678 extern int fio_options_parse(struct thread_data *, char **, int);
679 extern void fio_keywords_init(void);
680 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
681 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
682 extern void fio_fill_default_options(struct thread_data *);
683 extern int fio_show_option_help(const char *);
684 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
685 extern void fio_options_dup_and_init(struct option *);
686 extern void fio_options_mem_dupe(struct thread_data *);
687 extern void options_mem_dupe(void *data, struct fio_option *options);
688 extern void td_fill_rand_seeds(struct thread_data *);
689 extern void add_job_opts(const char **);
690 extern char *num2str(unsigned long, int, int, int, int);
691 extern int ioengine_load(struct thread_data *);
693 #define FIO_GETOPT_JOB 0x89000000
694 #define FIO_GETOPT_IOENGINE 0x98000000
695 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
700 extern void print_thread_status(void);
701 extern void print_status_init(int);
704 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
705 * will never back again. It may cycle between running/verififying/fsyncing.
706 * Once the thread reaches TD_EXITED, it is just waiting for the core to
723 extern void td_set_runstate(struct thread_data *, int);
724 #define TERMINATE_ALL (-1)
725 extern void fio_terminate_threads(int);
730 extern int __must_check fio_pin_memory(void);
731 extern void fio_unpin_memory(void);
732 extern int __must_check allocate_io_mem(struct thread_data *);
733 extern void free_io_mem(struct thread_data *);
736 * Reset stats after ramp time completes
738 extern void reset_all_stats(struct thread_data *);
743 #ifdef FIO_HAVE_BLKTRACE
744 extern int is_blktrace(const char *);
745 extern int load_blktrace(struct thread_data *, const char *);
749 * Mark unused variables passed to ops functions as unused, to silence gcc
751 #define fio_unused __attribute((__unused__))
752 #define fio_init __attribute__((constructor))
753 #define fio_exit __attribute__((destructor))
755 #define for_each_td(td, i) \
756 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
757 #define for_each_file(td, f, i) \
758 if ((td)->files_index) \
759 for ((i) = 0, (f) = (td)->files[0]; \
760 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
763 #define fio_assert(td, cond) do { \
766 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
767 td_set_runstate((td), TD_EXITED); \
768 (td)->error = EFAULT; \
773 static inline int fio_fill_issue_time(struct thread_data *td)
775 if (td->o.read_iolog_file ||
776 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
782 static inline int __should_check_rate(struct thread_data *td,
785 struct thread_options *o = &td->o;
788 * If some rate setting was given, we need to check it
790 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
791 o->rate_iops_min[ddir])
797 static inline int should_check_rate(struct thread_data *td,
798 uint64_t *bytes_done)
802 if (bytes_done[DDIR_READ])
803 ret |= __should_check_rate(td, DDIR_READ);
804 if (bytes_done[DDIR_WRITE])
805 ret |= __should_check_rate(td, DDIR_WRITE);
806 if (bytes_done[DDIR_TRIM])
807 ret |= __should_check_rate(td, DDIR_TRIM);
812 static inline unsigned int td_max_bs(struct thread_data *td)
816 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
817 return max(td->o.max_bs[DDIR_TRIM], max_bs);
820 static inline int is_power_of_2(unsigned int val)
822 return (val != 0 && ((val & (val - 1)) == 0));
826 * We currently only need to do locking if we have verifier threads
827 * accessing our internal structures too
829 static inline void td_io_u_lock(struct thread_data *td)
831 if (td->o.verify_async)
832 pthread_mutex_lock(&td->io_u_lock);
835 static inline void td_io_u_unlock(struct thread_data *td)
837 if (td->o.verify_async)
838 pthread_mutex_unlock(&td->io_u_lock);
841 static inline void td_io_u_free_notify(struct thread_data *td)
843 if (td->o.verify_async)
844 pthread_cond_signal(&td->free_cond);
847 extern const char *fio_get_arch_string(int);
848 extern const char *fio_get_os_string(int);
851 FIO_OUTPUT_TERSE = 0,
857 FIO_RAND_DIST_RANDOM = 0,
859 FIO_RAND_DIST_PARETO,
863 FIO_RAND_GEN_TAUSWORTHE = 0,