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 ddir_seq_nr;
113 unsigned int iodepth;
114 unsigned int iodepth_low;
115 unsigned int iodepth_batch;
116 unsigned int iodepth_batch_complete;
118 unsigned long long size;
119 unsigned int size_percent;
120 unsigned int fill_device;
121 unsigned long long file_size_low;
122 unsigned long long file_size_high;
123 unsigned long long start_offset;
125 unsigned int bs[DDIR_RWDIR_CNT];
126 unsigned int ba[DDIR_RWDIR_CNT];
127 unsigned int min_bs[DDIR_RWDIR_CNT];
128 unsigned int max_bs[DDIR_RWDIR_CNT];
129 struct bssplit *bssplit[DDIR_RWDIR_CNT];
130 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
132 int *ignore_error[ERROR_TYPE_CNT];
133 unsigned int ignore_error_nr[ERROR_TYPE_CNT];
134 unsigned int error_dump;
136 unsigned int nr_files;
137 unsigned int open_files;
138 enum file_lock_mode file_lock_mode;
140 unsigned int odirect;
141 unsigned int invalidate_cache;
142 unsigned int create_serialize;
143 unsigned int create_fsync;
144 unsigned int create_on_open;
145 unsigned int create_only;
146 unsigned int end_fsync;
147 unsigned int pre_read;
148 unsigned int sync_io;
150 unsigned int do_verify;
151 unsigned int verifysort;
152 unsigned int verifysort_nr;
153 unsigned int verify_interval;
154 unsigned int verify_offset;
155 char verify_pattern[MAX_PATTERN_SIZE];
156 unsigned int verify_pattern_bytes;
157 unsigned int verify_fatal;
158 unsigned int verify_dump;
159 unsigned int verify_async;
160 unsigned long long verify_backlog;
161 unsigned int verify_batch;
162 unsigned int experimental_verify;
163 unsigned int use_thread;
165 unsigned int do_disk_util;
166 unsigned int override_sync;
167 unsigned int rand_repeatable;
168 unsigned int use_os_rand;
169 unsigned int write_lat_log;
170 unsigned int write_bw_log;
171 unsigned int write_iops_log;
172 unsigned int log_avg_msec;
173 unsigned int norandommap;
174 unsigned int softrandommap;
175 unsigned int bs_unaligned;
176 unsigned int fsync_on_close;
178 unsigned int random_distribution;
182 unsigned int random_generator;
184 unsigned int hugepage_size;
185 unsigned int rw_min_bs;
186 unsigned int thinktime;
187 unsigned int thinktime_spin;
188 unsigned int thinktime_blocks;
189 unsigned int fsync_blocks;
190 unsigned int fdatasync_blocks;
191 unsigned int barrier_blocks;
192 unsigned long long start_delay;
193 unsigned long long timeout;
194 unsigned long long ramp_time;
195 unsigned int overwrite;
196 unsigned int bw_avg_time;
197 unsigned int iops_avg_time;
199 unsigned long long zone_range;
200 unsigned long long zone_size;
201 unsigned long long zone_skip;
202 enum fio_memtype mem_type;
203 unsigned int mem_align;
205 unsigned int max_latency;
207 unsigned int stonewall;
208 unsigned int new_group;
209 unsigned int numjobs;
210 os_cpu_mask_t cpumask;
211 unsigned int cpumask_set;
212 os_cpu_mask_t verify_cpumask;
213 unsigned int verify_cpumask_set;
214 #ifdef CONFIG_LIBNUMA
215 struct bitmask *numa_cpunodesmask;
216 unsigned int numa_cpumask_set;
217 unsigned short numa_mem_mode;
218 unsigned int numa_mem_prefer_node;
219 struct bitmask *numa_memnodesmask;
220 unsigned int numa_memmask_set;
223 unsigned int rwmixcycle;
224 unsigned int rwmix[2];
226 unsigned int file_service_type;
227 unsigned int group_reporting;
228 unsigned int fadvise_hint;
229 enum fio_fallocate_mode fallocate_mode;
230 unsigned int zero_buffers;
231 unsigned int refill_buffers;
232 unsigned int scramble_buffers;
233 unsigned int compress_percentage;
234 unsigned int compress_chunk;
235 unsigned int time_based;
236 unsigned int disable_lat;
237 unsigned int disable_clat;
238 unsigned int disable_slat;
239 unsigned int disable_bw;
240 unsigned int unified_rw_rep;
241 unsigned int gtod_reduce;
242 unsigned int gtod_cpu;
243 unsigned int gtod_offload;
244 enum fio_cs clocksource;
245 unsigned int no_stall;
246 unsigned int trim_percentage;
247 unsigned int trim_batch;
248 unsigned int trim_zero;
249 unsigned long long trim_backlog;
250 unsigned int clat_percentiles;
251 unsigned int percentile_precision; /* digits after decimal for percentiles */
252 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
254 char *read_iolog_file;
255 char *write_iolog_file;
259 char *replay_redirect;
262 * Pre-run and post-run shell
267 unsigned int rate[DDIR_RWDIR_CNT];
268 unsigned int ratemin[DDIR_RWDIR_CNT];
269 unsigned int ratecycle;
270 unsigned int rate_iops[DDIR_RWDIR_CNT];
271 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
276 * CPU "io" cycle burner
278 unsigned int cpuload;
279 unsigned int cpucycle;
284 enum error_type continue_on_error;
287 * Benchmark profile type
292 * blkio cgroup support
295 unsigned int cgroup_weight;
296 unsigned int cgroup_nodelete;
304 unsigned int flow_sleep;
306 unsigned long long offset_increment;
308 unsigned int sync_file_range;
312 TD_F_VER_BACKLOG = 1,
313 TD_F_TRIM_BACKLOG = 2,
315 TD_F_REFILL_BUFFERS = 8,
316 TD_F_SCRAMBLE_BUFFERS = 16,
318 TD_F_PROFILE_OPS = 64,
327 FIO_RAND_FILE_SIZE_OFF,
334 * This describes a single thread/process executing a fio job.
337 struct thread_options o;
340 char verror[FIO_VERROR_SIZE];
344 struct thread_stat ts;
346 struct io_log *slat_log;
347 struct io_log *clat_log;
348 struct io_log *lat_log;
349 struct io_log *bw_log;
350 struct io_log *iops_log;
352 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
353 struct timeval bw_sample_time;
355 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
356 struct timeval iops_sample_time;
358 volatile int update_rusage;
359 struct fio_mutex *rusage_sem;
360 struct rusage ru_start;
361 struct rusage ru_end;
363 struct fio_file **files;
364 unsigned char *file_locks;
365 unsigned int files_size;
366 unsigned int files_index;
367 unsigned int nr_open_files;
368 unsigned int nr_done_files;
369 unsigned int nr_normal_files;
371 unsigned int next_file;
372 os_random_state_t next_file_state;
373 struct frand_state __next_file_state;
380 size_t orig_buffer_size;
381 volatile int terminate;
382 volatile int runstate;
384 unsigned int ioprio_set;
385 unsigned int last_was_sync;
386 enum fio_ddir last_ddir;
396 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
399 os_random_state_t bsrange_state;
400 struct frand_state __bsrange_state;
403 os_random_state_t verify_state;
404 struct frand_state __verify_state;
407 os_random_state_t trim_state;
408 struct frand_state __trim_state;
411 struct frand_state buf_state;
413 unsigned int verify_batch;
414 unsigned int trim_batch;
419 * IO engine hooks, contains everything needed to submit an io_u
420 * to any of the available IO engines.
422 struct ioengine_ops *io_ops;
425 * Queue depth of io_u's that fio MIGHT do
427 unsigned int cur_depth;
430 * io_u's about to be committed
432 unsigned int io_u_queued;
435 * io_u's submitted but not completed yet
437 unsigned int io_u_in_flight;
440 * List of free and busy io_u's
442 struct flist_head io_u_freelist;
443 struct flist_head io_u_busylist;
444 struct flist_head io_u_requeues;
445 pthread_mutex_t io_u_lock;
446 pthread_cond_t free_cond;
449 * async verify offload
451 struct flist_head verify_list;
452 pthread_t *verify_threads;
453 unsigned int nr_verify_threads;
454 pthread_cond_t verify_cond;
455 int verify_thread_exit;
460 uint64_t rate_bps[DDIR_RWDIR_CNT];
461 long rate_pending_usleep[DDIR_RWDIR_CNT];
462 unsigned long rate_bytes[DDIR_RWDIR_CNT];
463 unsigned long rate_blocks[DDIR_RWDIR_CNT];
464 struct timeval lastrate[DDIR_RWDIR_CNT];
466 uint64_t total_io_size;
467 uint64_t fill_device_size;
469 unsigned long io_issues[DDIR_RWDIR_CNT];
470 uint64_t io_blocks[DDIR_RWDIR_CNT];
471 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
472 uint64_t io_bytes[DDIR_RWDIR_CNT];
473 uint64_t io_skip_bytes;
474 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
476 struct fio_mutex *mutex;
479 * State for random io, a bitmap of blocks done vs not done
482 os_random_state_t random_state;
483 struct frand_state __random_state;
486 struct timeval start; /* start of this loop */
487 struct timeval epoch; /* time job was started */
488 struct timeval last_issue;
489 struct timeval tv_cache;
490 unsigned int tv_cache_nr;
491 unsigned int tv_cache_mask;
492 unsigned int ramp_time_over;
495 * read/write mixed workload state
498 os_random_state_t rwmix_state;
499 struct frand_state __rwmix_state;
501 unsigned long rwmix_issues;
502 enum fio_ddir rwmix_ddir;
503 unsigned int ddir_seq_nr;
506 * IO history logs for verification. We use a tree for sorting,
507 * if we are overwriting. Otherwise just use a fifo.
509 struct rb_root io_hist_tree;
510 struct flist_head io_hist_list;
511 unsigned long io_hist_len;
516 struct flist_head io_log_list;
519 * For tracking/handling discards
521 struct flist_head trim_list;
522 unsigned long trim_entries;
524 struct flist_head next_rand_list;
527 * for fileservice, how often to switch to a new file
529 unsigned int file_service_nr;
530 unsigned int file_service_left;
531 struct fio_file *file_service_file;
533 unsigned int sync_file_range_nr;
536 * For generating file sizes
539 os_random_state_t file_size_state;
540 struct frand_state __file_size_state;
546 unsigned int total_err_count;
549 struct fio_flow *flow;
552 * Can be overloaded by profiles
554 struct prof_io_ops prof_io_ops;
559 * when should interactive ETA output be generated
567 #define __td_verror(td, err, msg, func) \
573 if (!(td)->first_error) \
574 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
578 #define td_clear_error(td) \
580 #define td_verror(td, err, func) \
581 __td_verror((td), (err), strerror((err)), (func))
582 #define td_vmsg(td, err, msg, func) \
583 __td_verror((td), (err), (msg), (func))
585 #define __fio_stringify_1(x) #x
586 #define __fio_stringify(x) __fio_stringify_1(x)
588 extern int exitall_on_terminate;
589 extern unsigned int thread_number;
590 extern unsigned int stat_number;
591 extern unsigned int nr_process, nr_thread;
594 extern int output_format;
595 extern int temp_stall_ts;
596 extern unsigned long long mlock_size;
597 extern uintptr_t page_mask, page_size;
598 extern int read_only;
599 extern int eta_print;
600 extern int eta_new_line;
601 extern unsigned long done_secs;
602 extern char *job_section;
603 extern int fio_gtod_offload;
604 extern int fio_gtod_cpu;
605 extern enum fio_cs fio_clock_source;
606 extern int fio_clock_source_set;
607 extern int warnings_fatal;
608 extern int terse_version;
609 extern int is_backend;
610 extern int nr_clients;
611 extern int log_syslog;
612 extern const char fio_version_string[];
614 extern struct thread_data *threads;
616 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
618 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
621 #define REAL_MAX_JOBS 2048
623 static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
626 return ERROR_TYPE_VERIFY_BIT;
627 if (ddir == DDIR_READ)
628 return ERROR_TYPE_READ_BIT;
629 return ERROR_TYPE_WRITE_BIT;
632 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
633 static inline int td_non_fatal_error(struct thread_data *td,
634 enum error_type_bit etype, int err)
637 if (!td->o.ignore_error[etype]) {
638 td->o.ignore_error[etype] = __NON_FATAL_ERR;
639 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
643 if (!(td->o.continue_on_error & (1 << etype)))
645 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
646 if (td->o.ignore_error[etype][i] == err)
651 static inline void update_error_count(struct thread_data *td, int err)
653 td->total_err_count++;
654 if (td->total_err_count == 1)
655 td->first_error = err;
658 static inline int should_fsync(struct thread_data *td)
660 if (td->last_was_sync)
662 if (td_write(td) || td_rw(td) || td->o.override_sync)
669 * Init/option functions
671 extern int __must_check parse_options(int, char **);
672 extern int parse_jobs_ini(char *, int, int);
673 extern int parse_cmd_line(int, char **);
674 extern int fio_backend(void);
675 extern void reset_fio_state(void);
676 extern void clear_io_state(struct thread_data *);
677 extern int fio_options_parse(struct thread_data *, char **, int);
678 extern void fio_keywords_init(void);
679 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
680 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
681 extern void fio_fill_default_options(struct thread_data *);
682 extern int fio_show_option_help(const char *);
683 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
684 extern void fio_options_dup_and_init(struct option *);
685 extern void fio_options_mem_dupe(struct thread_data *);
686 extern void options_mem_dupe(void *data, struct fio_option *options);
687 extern void td_fill_rand_seeds(struct thread_data *);
688 extern void add_job_opts(const char **);
689 extern char *num2str(unsigned long, int, int, int);
690 extern int ioengine_load(struct thread_data *);
692 #define FIO_GETOPT_JOB 0x89000000
693 #define FIO_GETOPT_IOENGINE 0x98000000
694 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
699 extern void print_thread_status(void);
700 extern void print_status_init(int);
703 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
704 * will never back again. It may cycle between running/verififying/fsyncing.
705 * Once the thread reaches TD_EXITED, it is just waiting for the core to
722 extern void td_set_runstate(struct thread_data *, int);
723 #define TERMINATE_ALL (-1)
724 extern void fio_terminate_threads(int);
729 extern int __must_check fio_pin_memory(void);
730 extern void fio_unpin_memory(void);
731 extern int __must_check allocate_io_mem(struct thread_data *);
732 extern void free_io_mem(struct thread_data *);
735 * Reset stats after ramp time completes
737 extern void reset_all_stats(struct thread_data *);
742 #ifdef FIO_HAVE_BLKTRACE
743 extern int is_blktrace(const char *);
744 extern int load_blktrace(struct thread_data *, const char *);
748 * Mark unused variables passed to ops functions as unused, to silence gcc
750 #define fio_unused __attribute((__unused__))
751 #define fio_init __attribute__((constructor))
752 #define fio_exit __attribute__((destructor))
754 #define for_each_td(td, i) \
755 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
756 #define for_each_file(td, f, i) \
757 if ((td)->files_index) \
758 for ((i) = 0, (f) = (td)->files[0]; \
759 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
762 #define fio_assert(td, cond) do { \
765 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
766 td_set_runstate((td), TD_EXITED); \
767 (td)->error = EFAULT; \
772 static inline int fio_fill_issue_time(struct thread_data *td)
774 if (td->o.read_iolog_file ||
775 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
781 static inline int __should_check_rate(struct thread_data *td,
784 struct thread_options *o = &td->o;
787 * If some rate setting was given, we need to check it
789 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
790 o->rate_iops_min[ddir])
796 static inline int should_check_rate(struct thread_data *td,
797 uint64_t *bytes_done)
801 if (bytes_done[DDIR_READ])
802 ret |= __should_check_rate(td, DDIR_READ);
803 if (bytes_done[DDIR_WRITE])
804 ret |= __should_check_rate(td, DDIR_WRITE);
805 if (bytes_done[DDIR_TRIM])
806 ret |= __should_check_rate(td, DDIR_TRIM);
811 static inline unsigned int td_max_bs(struct thread_data *td)
815 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
816 return max(td->o.max_bs[DDIR_TRIM], max_bs);
819 static inline int is_power_of_2(unsigned int val)
821 return (val != 0 && ((val & (val - 1)) == 0));
825 * We currently only need to do locking if we have verifier threads
826 * accessing our internal structures too
828 static inline void td_io_u_lock(struct thread_data *td)
830 if (td->o.verify_async)
831 pthread_mutex_lock(&td->io_u_lock);
834 static inline void td_io_u_unlock(struct thread_data *td)
836 if (td->o.verify_async)
837 pthread_mutex_unlock(&td->io_u_lock);
840 static inline void td_io_u_free_notify(struct thread_data *td)
842 if (td->o.verify_async)
843 pthread_cond_signal(&td->free_cond);
846 extern const char *fio_get_arch_string(int);
847 extern const char *fio_get_os_string(int);
850 FIO_OUTPUT_TERSE = 0,
856 FIO_RAND_DIST_RANDOM = 0,
858 FIO_RAND_DIST_PARETO,
862 FIO_RAND_GEN_TAUSWORTHE = 0,