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 {
107 enum td_ddir td_ddir;
109 unsigned int kb_base;
110 unsigned int ddir_seq_nr;
112 unsigned int iodepth;
113 unsigned int iodepth_low;
114 unsigned int iodepth_batch;
115 unsigned int iodepth_batch_complete;
117 unsigned long long size;
118 unsigned int size_percent;
119 unsigned int fill_device;
120 unsigned long long file_size_low;
121 unsigned long long file_size_high;
122 unsigned long long start_offset;
124 unsigned int bs[DDIR_RWDIR_CNT];
125 unsigned int ba[DDIR_RWDIR_CNT];
126 unsigned int min_bs[DDIR_RWDIR_CNT];
127 unsigned int max_bs[DDIR_RWDIR_CNT];
128 struct bssplit *bssplit[DDIR_RWDIR_CNT];
129 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
131 int *ignore_error[ERROR_TYPE_CNT];
132 unsigned int ignore_error_nr[ERROR_TYPE_CNT];
133 unsigned int error_dump;
135 unsigned int nr_files;
136 unsigned int open_files;
137 enum file_lock_mode file_lock_mode;
139 unsigned int odirect;
140 unsigned int invalidate_cache;
141 unsigned int create_serialize;
142 unsigned int create_fsync;
143 unsigned int create_on_open;
144 unsigned int create_only;
145 unsigned int end_fsync;
146 unsigned int pre_read;
147 unsigned int sync_io;
149 unsigned int do_verify;
150 unsigned int verifysort;
151 unsigned int verifysort_nr;
152 unsigned int verify_interval;
153 unsigned int verify_offset;
154 char verify_pattern[MAX_PATTERN_SIZE];
155 unsigned int verify_pattern_bytes;
156 unsigned int verify_fatal;
157 unsigned int verify_dump;
158 unsigned int verify_async;
159 unsigned long long verify_backlog;
160 unsigned int verify_batch;
161 unsigned int experimental_verify;
162 unsigned int use_thread;
164 unsigned int do_disk_util;
165 unsigned int override_sync;
166 unsigned int rand_repeatable;
167 unsigned int use_os_rand;
168 unsigned int write_lat_log;
169 unsigned int write_bw_log;
170 unsigned int write_iops_log;
171 unsigned int log_avg_msec;
172 unsigned int norandommap;
173 unsigned int softrandommap;
174 unsigned int bs_unaligned;
175 unsigned int fsync_on_close;
177 unsigned int random_distribution;
181 unsigned int random_generator;
183 unsigned int hugepage_size;
184 unsigned int rw_min_bs;
185 unsigned int thinktime;
186 unsigned int thinktime_spin;
187 unsigned int thinktime_blocks;
188 unsigned int fsync_blocks;
189 unsigned int fdatasync_blocks;
190 unsigned int barrier_blocks;
191 unsigned long long start_delay;
192 unsigned long long timeout;
193 unsigned long long ramp_time;
194 unsigned int overwrite;
195 unsigned int bw_avg_time;
196 unsigned int iops_avg_time;
198 unsigned long long zone_range;
199 unsigned long long zone_size;
200 unsigned long long zone_skip;
201 enum fio_memtype mem_type;
202 unsigned int mem_align;
204 unsigned int max_latency;
206 unsigned int stonewall;
207 unsigned int new_group;
208 unsigned int numjobs;
209 os_cpu_mask_t cpumask;
210 unsigned int cpumask_set;
211 os_cpu_mask_t verify_cpumask;
212 unsigned int verify_cpumask_set;
213 #ifdef CONFIG_LIBNUMA
214 struct bitmask *numa_cpunodesmask;
215 unsigned int numa_cpumask_set;
216 unsigned short numa_mem_mode;
217 unsigned int numa_mem_prefer_node;
218 struct bitmask *numa_memnodesmask;
219 unsigned int numa_memmask_set;
222 unsigned int rwmixcycle;
223 unsigned int rwmix[2];
225 unsigned int file_service_type;
226 unsigned int group_reporting;
227 unsigned int fadvise_hint;
228 enum fio_fallocate_mode fallocate_mode;
229 unsigned int zero_buffers;
230 unsigned int refill_buffers;
231 unsigned int scramble_buffers;
232 unsigned int compress_percentage;
233 unsigned int compress_chunk;
234 unsigned int time_based;
235 unsigned int disable_lat;
236 unsigned int disable_clat;
237 unsigned int disable_slat;
238 unsigned int disable_bw;
239 unsigned int unified_rw_rep;
240 unsigned int gtod_reduce;
241 unsigned int gtod_cpu;
242 unsigned int gtod_offload;
243 enum fio_cs clocksource;
244 unsigned int no_stall;
245 unsigned int trim_percentage;
246 unsigned int trim_batch;
247 unsigned int trim_zero;
248 unsigned long long trim_backlog;
249 unsigned int clat_percentiles;
250 unsigned int percentile_precision; /* digits after decimal for percentiles */
251 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
253 char *read_iolog_file;
254 char *write_iolog_file;
258 char *replay_redirect;
261 * Pre-run and post-run shell
266 unsigned int rate[DDIR_RWDIR_CNT];
267 unsigned int ratemin[DDIR_RWDIR_CNT];
268 unsigned int ratecycle;
269 unsigned int rate_iops[DDIR_RWDIR_CNT];
270 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
275 * CPU "io" cycle burner
277 unsigned int cpuload;
278 unsigned int cpucycle;
283 enum error_type continue_on_error;
286 * Benchmark profile type
291 * blkio cgroup support
294 unsigned int cgroup_weight;
295 unsigned int cgroup_nodelete;
303 unsigned int flow_sleep;
305 unsigned long long offset_increment;
307 unsigned int sync_file_range;
311 TD_F_VER_BACKLOG = 1,
312 TD_F_TRIM_BACKLOG = 2,
314 TD_F_REFILL_BUFFERS = 8,
315 TD_F_SCRAMBLE_BUFFERS = 16,
317 TD_F_PROFILE_OPS = 64,
326 FIO_RAND_FILE_SIZE_OFF,
333 * This describes a single thread/process executing a fio job.
336 struct thread_options o;
339 char verror[FIO_VERROR_SIZE];
343 struct thread_stat ts;
345 struct io_log *slat_log;
346 struct io_log *clat_log;
347 struct io_log *lat_log;
348 struct io_log *bw_log;
349 struct io_log *iops_log;
351 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
352 struct timeval bw_sample_time;
354 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
355 struct timeval iops_sample_time;
357 volatile int update_rusage;
358 struct fio_mutex *rusage_sem;
359 struct rusage ru_start;
360 struct rusage ru_end;
362 struct fio_file **files;
363 unsigned char *file_locks;
364 unsigned int files_size;
365 unsigned int files_index;
366 unsigned int nr_open_files;
367 unsigned int nr_done_files;
368 unsigned int nr_normal_files;
370 unsigned int next_file;
371 os_random_state_t next_file_state;
372 struct frand_state __next_file_state;
379 size_t orig_buffer_size;
380 volatile int terminate;
381 volatile int runstate;
383 unsigned int ioprio_set;
384 unsigned int last_was_sync;
385 enum fio_ddir last_ddir;
395 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
398 os_random_state_t bsrange_state;
399 struct frand_state __bsrange_state;
402 os_random_state_t verify_state;
403 struct frand_state __verify_state;
406 os_random_state_t trim_state;
407 struct frand_state __trim_state;
410 struct frand_state buf_state;
412 unsigned int verify_batch;
413 unsigned int trim_batch;
418 * IO engine hooks, contains everything needed to submit an io_u
419 * to any of the available IO engines.
421 struct ioengine_ops *io_ops;
424 * Queue depth of io_u's that fio MIGHT do
426 unsigned int cur_depth;
429 * io_u's about to be committed
431 unsigned int io_u_queued;
434 * io_u's submitted but not completed yet
436 unsigned int io_u_in_flight;
439 * List of free and busy io_u's
441 struct flist_head io_u_freelist;
442 struct flist_head io_u_busylist;
443 struct flist_head io_u_requeues;
444 pthread_mutex_t io_u_lock;
445 pthread_cond_t free_cond;
448 * async verify offload
450 struct flist_head verify_list;
451 pthread_t *verify_threads;
452 unsigned int nr_verify_threads;
453 pthread_cond_t verify_cond;
454 int verify_thread_exit;
459 uint64_t rate_bps[DDIR_RWDIR_CNT];
460 long rate_pending_usleep[DDIR_RWDIR_CNT];
461 unsigned long rate_bytes[DDIR_RWDIR_CNT];
462 unsigned long rate_blocks[DDIR_RWDIR_CNT];
463 struct timeval lastrate[DDIR_RWDIR_CNT];
465 uint64_t total_io_size;
466 uint64_t fill_device_size;
468 unsigned long io_issues[DDIR_RWDIR_CNT];
469 uint64_t io_blocks[DDIR_RWDIR_CNT];
470 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
471 uint64_t io_bytes[DDIR_RWDIR_CNT];
472 uint64_t io_skip_bytes;
473 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
475 struct fio_mutex *mutex;
478 * State for random io, a bitmap of blocks done vs not done
481 os_random_state_t random_state;
482 struct frand_state __random_state;
485 struct timeval start; /* start of this loop */
486 struct timeval epoch; /* time job was started */
487 struct timeval last_issue;
488 struct timeval tv_cache;
489 unsigned int tv_cache_nr;
490 unsigned int tv_cache_mask;
491 unsigned int ramp_time_over;
494 * read/write mixed workload state
497 os_random_state_t rwmix_state;
498 struct frand_state __rwmix_state;
500 unsigned long rwmix_issues;
501 enum fio_ddir rwmix_ddir;
502 unsigned int ddir_seq_nr;
505 * IO history logs for verification. We use a tree for sorting,
506 * if we are overwriting. Otherwise just use a fifo.
508 struct rb_root io_hist_tree;
509 struct flist_head io_hist_list;
510 unsigned long io_hist_len;
515 struct flist_head io_log_list;
518 * For tracking/handling discards
520 struct flist_head trim_list;
521 unsigned long trim_entries;
523 struct flist_head next_rand_list;
526 * for fileservice, how often to switch to a new file
528 unsigned int file_service_nr;
529 unsigned int file_service_left;
530 struct fio_file *file_service_file;
532 unsigned int sync_file_range_nr;
535 * For generating file sizes
538 os_random_state_t file_size_state;
539 struct frand_state __file_size_state;
545 unsigned int total_err_count;
548 struct fio_flow *flow;
551 * Can be overloaded by profiles
553 struct prof_io_ops prof_io_ops;
558 * when should interactive ETA output be generated
566 #define __td_verror(td, err, msg, func) \
572 if (!(td)->first_error) \
573 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
577 #define td_clear_error(td) \
579 #define td_verror(td, err, func) \
580 __td_verror((td), (err), strerror((err)), (func))
581 #define td_vmsg(td, err, msg, func) \
582 __td_verror((td), (err), (msg), (func))
584 #define __fio_stringify_1(x) #x
585 #define __fio_stringify(x) __fio_stringify_1(x)
587 extern int exitall_on_terminate;
588 extern unsigned int thread_number;
589 extern unsigned int stat_number;
590 extern unsigned int nr_process, nr_thread;
593 extern int output_format;
594 extern int temp_stall_ts;
595 extern unsigned long long mlock_size;
596 extern uintptr_t page_mask, page_size;
597 extern int read_only;
598 extern int eta_print;
599 extern int eta_new_line;
600 extern unsigned long done_secs;
601 extern char *job_section;
602 extern int fio_gtod_offload;
603 extern int fio_gtod_cpu;
604 extern enum fio_cs fio_clock_source;
605 extern int fio_clock_source_set;
606 extern int warnings_fatal;
607 extern int terse_version;
608 extern int is_backend;
609 extern int nr_clients;
610 extern int log_syslog;
611 extern const char fio_version_string[];
613 extern struct thread_data *threads;
615 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
617 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
620 #define REAL_MAX_JOBS 2048
622 static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
625 return ERROR_TYPE_VERIFY_BIT;
626 if (ddir == DDIR_READ)
627 return ERROR_TYPE_READ_BIT;
628 return ERROR_TYPE_WRITE_BIT;
631 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
632 static inline int td_non_fatal_error(struct thread_data *td,
633 enum error_type_bit etype, int err)
636 if (!td->o.ignore_error[etype]) {
637 td->o.ignore_error[etype] = __NON_FATAL_ERR;
638 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
642 if (!(td->o.continue_on_error & (1 << etype)))
644 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
645 if (td->o.ignore_error[etype][i] == err)
650 static inline void update_error_count(struct thread_data *td, int err)
652 td->total_err_count++;
653 if (td->total_err_count == 1)
654 td->first_error = err;
657 static inline int should_fsync(struct thread_data *td)
659 if (td->last_was_sync)
661 if (td_write(td) || td_rw(td) || td->o.override_sync)
668 * Init/option functions
670 extern int __must_check parse_options(int, char **);
671 extern int parse_jobs_ini(char *, int, int);
672 extern int parse_cmd_line(int, char **);
673 extern int fio_backend(void);
674 extern void reset_fio_state(void);
675 extern void clear_io_state(struct thread_data *);
676 extern int fio_options_parse(struct thread_data *, char **, int);
677 extern void fio_keywords_init(void);
678 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
679 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
680 extern void fio_fill_default_options(struct thread_data *);
681 extern int fio_show_option_help(const char *);
682 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
683 extern void fio_options_dup_and_init(struct option *);
684 extern void fio_options_mem_dupe(struct thread_data *);
685 extern void options_mem_dupe(void *data, struct fio_option *options);
686 extern void td_fill_rand_seeds(struct thread_data *);
687 extern void add_job_opts(const char **);
688 extern char *num2str(unsigned long, int, int, int);
689 extern int ioengine_load(struct thread_data *);
691 #define FIO_GETOPT_JOB 0x89000000
692 #define FIO_GETOPT_IOENGINE 0x98000000
693 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
698 extern void print_thread_status(void);
699 extern void print_status_init(int);
702 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
703 * will never back again. It may cycle between running/verififying/fsyncing.
704 * Once the thread reaches TD_EXITED, it is just waiting for the core to
721 extern void td_set_runstate(struct thread_data *, int);
722 #define TERMINATE_ALL (-1)
723 extern void fio_terminate_threads(int);
728 extern int __must_check fio_pin_memory(void);
729 extern void fio_unpin_memory(void);
730 extern int __must_check allocate_io_mem(struct thread_data *);
731 extern void free_io_mem(struct thread_data *);
734 * Reset stats after ramp time completes
736 extern void reset_all_stats(struct thread_data *);
741 #ifdef FIO_HAVE_BLKTRACE
742 extern int is_blktrace(const char *);
743 extern int load_blktrace(struct thread_data *, const char *);
747 * Mark unused variables passed to ops functions as unused, to silence gcc
749 #define fio_unused __attribute((__unused__))
750 #define fio_init __attribute__((constructor))
751 #define fio_exit __attribute__((destructor))
753 #define for_each_td(td, i) \
754 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
755 #define for_each_file(td, f, i) \
756 if ((td)->files_index) \
757 for ((i) = 0, (f) = (td)->files[0]; \
758 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
761 #define fio_assert(td, cond) do { \
764 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
765 td_set_runstate((td), TD_EXITED); \
766 (td)->error = EFAULT; \
771 static inline int fio_fill_issue_time(struct thread_data *td)
773 if (td->o.read_iolog_file ||
774 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
780 static inline int __should_check_rate(struct thread_data *td,
783 struct thread_options *o = &td->o;
786 * If some rate setting was given, we need to check it
788 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
789 o->rate_iops_min[ddir])
795 static inline int should_check_rate(struct thread_data *td,
796 uint64_t *bytes_done)
800 if (bytes_done[DDIR_READ])
801 ret |= __should_check_rate(td, DDIR_READ);
802 if (bytes_done[DDIR_WRITE])
803 ret |= __should_check_rate(td, DDIR_WRITE);
804 if (bytes_done[DDIR_TRIM])
805 ret |= __should_check_rate(td, DDIR_TRIM);
810 static inline unsigned int td_max_bs(struct thread_data *td)
814 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
815 return max(td->o.max_bs[DDIR_TRIM], max_bs);
818 static inline int is_power_of_2(unsigned int val)
820 return (val != 0 && ((val & (val - 1)) == 0));
824 * We currently only need to do locking if we have verifier threads
825 * accessing our internal structures too
827 static inline void td_io_u_lock(struct thread_data *td)
829 if (td->o.verify_async)
830 pthread_mutex_lock(&td->io_u_lock);
833 static inline void td_io_u_unlock(struct thread_data *td)
835 if (td->o.verify_async)
836 pthread_mutex_unlock(&td->io_u_lock);
839 static inline void td_io_u_free_notify(struct thread_data *td)
841 if (td->o.verify_async)
842 pthread_cond_signal(&td->free_cond);
845 extern const char *fio_get_arch_string(int);
846 extern const char *fio_get_os_string(int);
849 FIO_OUTPUT_TERSE = 0,
855 FIO_RAND_DIST_RANDOM = 0,
857 FIO_RAND_DIST_PARETO,
861 FIO_RAND_GEN_TAUSWORTHE = 0,