8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
35 #include "oslib/getopt.h"
37 #include "lib/rbtree.h"
42 #include "io_u_queue.h"
43 #include "workqueue.h"
45 #ifdef CONFIG_SOLARISAIO
46 #include <sys/asynch.h>
50 #include <linux/mempolicy.h>
54 * "local" is pseudo-policy
56 #define MPOL_LOCAL MPOL_MAX
60 * offset generator types
68 TD_F_VER_BACKLOG = 1U << 0,
69 TD_F_TRIM_BACKLOG = 1U << 1,
70 TD_F_READ_IOLOG = 1U << 2,
71 TD_F_REFILL_BUFFERS = 1U << 3,
72 TD_F_SCRAMBLE_BUFFERS = 1U << 4,
73 TD_F_VER_NONE = 1U << 5,
74 TD_F_PROFILE_OPS = 1U << 6,
75 TD_F_COMPRESS = 1U << 7,
77 TD_F_COMPRESS_LOG = 1U << 9,
78 TD_F_VSTATE_SAVED = 1U << 10,
79 TD_F_NEED_LOCK = 1U << 11,
80 TD_F_CHILD = 1U << 12,
81 TD_F_NO_PROGRESS = 1U << 13,
82 TD_F_REGROW_LOGS = 1U << 14,
91 FIO_RAND_FILE_SIZE_OFF,
94 FIO_RAND_SEQ_RAND_READ_OFF,
95 FIO_RAND_SEQ_RAND_WRITE_OFF,
96 FIO_RAND_SEQ_RAND_TRIM_OFF,
108 RATE_PROCESS_LINEAR = 0,
109 RATE_PROCESS_POISSON = 1,
113 * Per-thread/process specific data. Only used for the network client
117 void sk_out_assign(struct sk_out *);
118 void sk_out_drop(void);
120 struct zone_split_index {
122 uint8_t size_perc_prev;
126 * For steady state detection
128 struct steadystate_data {
130 unsigned long long dur;
131 unsigned long long ramp_time;
132 bool (*evaluate)(unsigned long, unsigned long, struct thread_data *);
143 unsigned long *iops_data;
144 unsigned long *bw_data;
150 unsigned long long sum_y;
151 unsigned long long sum_x;
152 unsigned long long sum_x_sq;
153 unsigned long long sum_xy;
154 unsigned long long oldest_y;
156 struct timeval prev_time;
157 unsigned long long prev_iops;
158 unsigned long long prev_bytes;
163 * This describes a single thread/process executing a fio job.
166 struct flist_head opt_list;
168 struct thread_options o;
171 unsigned int thread_number;
172 unsigned int subjob_number;
173 unsigned int groupid;
174 struct thread_stat ts;
178 struct io_log *slat_log;
179 struct io_log *clat_log;
180 struct io_log *clat_hist_log;
181 struct io_log *lat_log;
182 struct io_log *bw_log;
183 struct io_log *iops_log;
185 struct workqueue log_compress_wq;
187 struct thread_data *parent;
189 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
190 struct timeval bw_sample_time;
192 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
193 struct timeval iops_sample_time;
195 volatile int update_rusage;
196 struct fio_mutex *rusage_sem;
197 struct rusage ru_start;
198 struct rusage ru_end;
200 struct fio_file **files;
201 unsigned char *file_locks;
202 unsigned int files_size;
203 unsigned int files_index;
204 unsigned int nr_open_files;
205 unsigned int nr_done_files;
206 unsigned int nr_normal_files;
208 unsigned int next_file;
209 struct frand_state next_file_state;
212 struct zipf_state next_file_zipf;
213 struct gauss_state next_file_gauss;
226 size_t orig_buffer_size;
227 volatile int terminate;
228 volatile int runstate;
229 unsigned int last_was_sync;
230 enum fio_ddir last_ddir;
239 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
241 struct frand_state bsrange_state;
242 struct frand_state verify_state;
243 struct frand_state trim_state;
244 struct frand_state delay_state;
246 struct frand_state buf_state;
247 struct frand_state buf_state_prev;
248 struct frand_state dedupe_state;
249 struct frand_state zone_state;
251 struct zone_split_index **zone_state_index;
253 unsigned int verify_batch;
254 unsigned int trim_batch;
256 struct thread_io_list *vstate;
261 * IO engine hooks, contains everything needed to submit an io_u
262 * to any of the available IO engines.
264 struct ioengine_ops *io_ops;
267 * IO engine private data and dlhandle.
270 void *io_ops_dlhandle;
273 * Queue depth of io_u's that fio MIGHT do
275 unsigned int cur_depth;
278 * io_u's about to be committed
280 unsigned int io_u_queued;
283 * io_u's submitted but not completed yet
285 unsigned int io_u_in_flight;
288 * List of free and busy io_u's
290 struct io_u_ring io_u_requeues;
291 struct io_u_queue io_u_freelist;
292 struct io_u_queue io_u_all;
293 pthread_mutex_t io_u_lock;
294 pthread_cond_t free_cond;
297 * async verify offload
299 struct flist_head verify_list;
300 pthread_t *verify_threads;
301 unsigned int nr_verify_threads;
302 pthread_cond_t verify_cond;
303 int verify_thread_exit;
308 uint64_t rate_bps[DDIR_RWDIR_CNT];
309 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
310 unsigned long rate_bytes[DDIR_RWDIR_CNT];
311 unsigned long rate_blocks[DDIR_RWDIR_CNT];
312 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
313 struct timeval lastrate[DDIR_RWDIR_CNT];
315 struct frand_state poisson_state;
318 * Enforced rate submission/completion workqueue
320 struct workqueue io_wq;
322 uint64_t total_io_size;
323 uint64_t fill_device_size;
328 uint64_t io_issues[DDIR_RWDIR_CNT];
329 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
335 uint64_t io_blocks[DDIR_RWDIR_CNT];
336 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
337 uint64_t io_bytes[DDIR_RWDIR_CNT];
338 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
339 uint64_t io_skip_bytes;
341 struct fio_mutex *mutex;
342 uint64_t bytes_done[DDIR_RWDIR_CNT];
345 * State for random io, a bitmap of blocks done vs not done
347 struct frand_state random_state;
349 struct timeval start; /* start of this loop */
350 struct timeval epoch; /* time job was started */
351 struct timeval last_issue;
353 struct timeval tv_cache;
354 struct timeval terminate_time;
355 unsigned int tv_cache_nr;
356 unsigned int tv_cache_mask;
357 unsigned int ramp_time_over;
360 * Time since last latency_window was started
362 struct timeval latency_ts;
363 unsigned int latency_qd;
364 unsigned int latency_qd_high;
365 unsigned int latency_qd_low;
366 unsigned int latency_failed;
367 uint64_t latency_ios;
371 * read/write mixed workload state
373 struct frand_state rwmix_state;
374 unsigned long rwmix_issues;
375 enum fio_ddir rwmix_ddir;
376 unsigned int ddir_seq_nr;
379 * rand/seq mixed workload state
381 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
384 * IO history logs for verification. We use a tree for sorting,
385 * if we are overwriting. Otherwise just use a fifo.
387 struct rb_root io_hist_tree;
388 struct flist_head io_hist_list;
389 unsigned long io_hist_len;
394 struct flist_head io_log_list;
397 * For tracking/handling discards
399 struct flist_head trim_list;
400 unsigned long trim_entries;
402 struct flist_head next_rand_list;
405 * for fileservice, how often to switch to a new file
407 unsigned int file_service_nr;
408 unsigned int file_service_left;
409 struct fio_file *file_service_file;
411 unsigned int sync_file_range_nr;
414 * For generating file sizes
416 struct frand_state file_size_state;
421 unsigned int total_err_count;
424 struct fio_flow *flow;
427 * Can be overloaded by profiles
429 struct prof_io_ops prof_io_ops;
434 struct steadystate_data ss;
436 char verror[FIO_VERROR_SIZE];
440 * when should interactive ETA output be generated
448 #define __td_verror(td, err, msg, func) \
450 unsigned int ____e = (err); \
453 (td)->error = ____e; \
454 if (!(td)->first_error) \
455 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
459 #define td_clear_error(td) do { \
462 (td)->parent->error = 0; \
465 #define td_verror(td, err, func) do { \
466 __td_verror((td), (err), strerror((err)), (func)); \
468 __td_verror((td)->parent, (err), strerror((err)), (func)); \
471 #define td_vmsg(td, err, msg, func) do { \
472 __td_verror((td), (err), (msg), (func)); \
474 __td_verror((td)->parent, (err), (msg), (func)); \
477 #define __fio_stringify_1(x) #x
478 #define __fio_stringify(x) __fio_stringify_1(x)
480 extern int exitall_on_terminate;
481 extern unsigned int thread_number;
482 extern unsigned int stat_number;
485 extern int output_format;
486 extern int append_terse_output;
487 extern int temp_stall_ts;
488 extern uintptr_t page_mask, page_size;
489 extern int read_only;
490 extern int eta_print;
491 extern int eta_new_line;
492 extern unsigned long done_secs;
493 extern char *job_section;
494 extern int fio_gtod_offload;
495 extern int fio_gtod_cpu;
496 extern enum fio_cs fio_clock_source;
497 extern int fio_clock_source_set;
498 extern int warnings_fatal;
499 extern int terse_version;
500 extern int is_backend;
501 extern int nr_clients;
502 extern int log_syslog;
503 extern int status_interval;
504 extern const char fio_version_string[];
505 extern char *trigger_file;
506 extern char *trigger_cmd;
507 extern char *trigger_remote_cmd;
508 extern long long trigger_timeout;
509 extern char *aux_path;
511 extern bool steadystate;
512 #define STEADYSTATE_MSEC (1000)
514 extern struct thread_data *threads;
516 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
518 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
521 #define REAL_MAX_JOBS 2048
523 static inline int should_fsync(struct thread_data *td)
525 if (td->last_was_sync)
527 if (td_write(td) || td_rw(td) || td->o.override_sync)
534 * Init/option functions
536 extern int __must_check fio_init_options(void);
537 extern int __must_check parse_options(int, char **);
538 extern int parse_jobs_ini(char *, int, int, int);
539 extern int parse_cmd_line(int, char **, int);
540 extern int fio_backend(struct sk_out *);
541 extern void reset_fio_state(void);
542 extern void clear_io_state(struct thread_data *, int);
543 extern int fio_options_parse(struct thread_data *, char **, int);
544 extern void fio_keywords_init(void);
545 extern void fio_keywords_exit(void);
546 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
547 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
548 extern void fio_fill_default_options(struct thread_data *);
549 extern int fio_show_option_help(const char *);
550 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
551 extern void fio_options_dup_and_init(struct option *);
552 extern void fio_options_mem_dupe(struct thread_data *);
553 extern void options_mem_dupe(void *data, struct fio_option *options);
554 extern void td_fill_rand_seeds(struct thread_data *);
555 extern void td_fill_verify_state_seed(struct thread_data *);
556 extern void add_job_opts(const char **, int);
557 extern char *num2str(uint64_t, int, int, int, int);
558 extern int ioengine_load(struct thread_data *);
559 extern int parse_dryrun(void);
560 extern int fio_running_or_pending_io_threads(void);
561 extern int fio_set_fd_nonblocking(int, const char *);
562 extern void sig_show_status(int sig);
563 extern struct thread_data *get_global_options(void);
565 extern uintptr_t page_mask;
566 extern uintptr_t page_size;
567 extern int initialize_fio(char *envp[]);
568 extern void deinitialize_fio(void);
570 #define FIO_GETOPT_JOB 0x89000000
571 #define FIO_GETOPT_IOENGINE 0x98000000
572 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
577 extern void print_thread_status(void);
578 extern void print_status_init(int);
579 extern char *fio_uint_to_kmg(unsigned int val);
582 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
583 * will never back again. It may cycle between running/verififying/fsyncing.
584 * Once the thread reaches TD_EXITED, it is just waiting for the core to
604 #define TD_ENG_FLAG_SHIFT 16
605 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
607 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
609 return (td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK;
612 static inline void td_set_ioengine_flags(struct thread_data *td)
614 td->flags |= (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
617 static inline bool td_ioengine_flagged(struct thread_data *td, unsigned int val)
619 return ((td->flags >> TD_ENG_FLAG_SHIFT) & val) != 0;
622 extern void td_set_runstate(struct thread_data *, int);
623 extern int td_bump_runstate(struct thread_data *, int);
624 extern void td_restore_runstate(struct thread_data *, int);
625 extern const char *runstate_to_name(int runstate);
628 * Allow 60 seconds for a job to quit on its own, otherwise reap with
631 #define FIO_REAP_TIMEOUT 60
633 #define TERMINATE_ALL (-1U)
634 extern void fio_terminate_threads(unsigned int);
635 extern void fio_mark_td_terminate(struct thread_data *);
640 extern int __must_check fio_pin_memory(struct thread_data *);
641 extern void fio_unpin_memory(struct thread_data *);
642 extern int __must_check allocate_io_mem(struct thread_data *);
643 extern void free_io_mem(struct thread_data *);
644 extern void free_threads_shm(void);
647 * Reset stats after ramp time completes
649 extern void reset_all_stats(struct thread_data *);
654 #ifdef FIO_HAVE_BLKTRACE
655 extern int is_blktrace(const char *, int *);
656 extern int load_blktrace(struct thread_data *, const char *, int);
659 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
660 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
661 struct timeval *comp_time);
664 * Latency target helpers
666 extern void lat_target_check(struct thread_data *);
667 extern void lat_target_init(struct thread_data *);
668 extern void lat_target_reset(struct thread_data *);
670 #define for_each_td(td, i) \
671 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
672 #define for_each_file(td, f, i) \
673 if ((td)->files_index) \
674 for ((i) = 0, (f) = (td)->files[0]; \
675 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
678 #define fio_assert(td, cond) do { \
681 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
682 td_set_runstate((td), TD_EXITED); \
683 (td)->error = EFAULT; \
688 static inline int fio_fill_issue_time(struct thread_data *td)
690 if (td->o.read_iolog_file ||
691 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
697 static inline int __should_check_rate(struct thread_data *td,
700 struct thread_options *o = &td->o;
703 * If some rate setting was given, we need to check it
705 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
706 o->rate_iops_min[ddir])
712 static inline int should_check_rate(struct thread_data *td)
716 if (td->bytes_done[DDIR_READ])
717 ret |= __should_check_rate(td, DDIR_READ);
718 if (td->bytes_done[DDIR_WRITE])
719 ret |= __should_check_rate(td, DDIR_WRITE);
720 if (td->bytes_done[DDIR_TRIM])
721 ret |= __should_check_rate(td, DDIR_TRIM);
726 static inline unsigned int td_max_bs(struct thread_data *td)
730 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
731 return max(td->o.max_bs[DDIR_TRIM], max_bs);
734 static inline unsigned int td_min_bs(struct thread_data *td)
738 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
739 return min(td->o.min_bs[DDIR_TRIM], min_bs);
742 static inline bool td_async_processing(struct thread_data *td)
744 return (td->flags & TD_F_NEED_LOCK) != 0;
748 * We currently only need to do locking if we have verifier threads
749 * accessing our internal structures too
751 static inline void td_io_u_lock(struct thread_data *td)
753 if (td_async_processing(td))
754 pthread_mutex_lock(&td->io_u_lock);
757 static inline void td_io_u_unlock(struct thread_data *td)
759 if (td_async_processing(td))
760 pthread_mutex_unlock(&td->io_u_lock);
763 static inline void td_io_u_free_notify(struct thread_data *td)
765 if (td_async_processing(td))
766 pthread_cond_signal(&td->free_cond);
769 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
772 if (!td_async_processing(td))
775 __sync_fetch_and_and(flags, ~value);
778 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
781 if (!td_async_processing(td))
784 __sync_fetch_and_or(flags, value);
787 extern const char *fio_get_arch_string(int);
788 extern const char *fio_get_os_string(int);
791 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
792 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
796 __FIO_OUTPUT_TERSE = 0,
797 __FIO_OUTPUT_JSON = 1,
798 __FIO_OUTPUT_NORMAL = 2,
799 __FIO_OUTPUT_JSON_PLUS = 3,
802 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
803 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
804 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
805 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
809 FIO_RAND_DIST_RANDOM = 0,
811 FIO_RAND_DIST_PARETO,
816 #define FIO_DEF_ZIPF 1.1
817 #define FIO_DEF_PARETO 0.2
820 FIO_RAND_GEN_TAUSWORTHE = 0,
822 FIO_RAND_GEN_TAUSWORTHE64,
831 FIO_STEADYSTATE_IOPS = 0,
832 FIO_STEADYSTATE_IOPS_SLOPE,
834 FIO_STEADYSTATE_BW_SLOPE,
838 extern void exec_trigger(const char *);
839 extern void check_trigger_file(void);