8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
27 #include "ioengines.h"
35 #include "oslib/getopt.h"
37 #include "lib/rbtree.h"
38 #include "lib/num2str.h"
44 #include "io_u_queue.h"
45 #include "workqueue.h"
46 #include "steadystate.h"
47 #include "lib/nowarn_snprintf.h"
49 #ifdef CONFIG_SOLARISAIO
50 #include <sys/asynch.h>
54 #include <linux/mempolicy.h>
58 * "local" is pseudo-policy
72 * offset generator types
80 __TD_F_VER_BACKLOG = 0,
83 __TD_F_REFILL_BUFFERS,
84 __TD_F_SCRAMBLE_BUFFERS,
97 __TD_F_LAST, /* not a real bit, keep last */
101 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
102 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
103 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
104 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
105 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
106 TD_F_DO_VERIFY = 1U << __TD_F_DO_VERIFY,
107 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
108 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
109 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
110 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
111 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
112 TD_F_CHILD = 1U << __TD_F_CHILD,
113 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
114 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
115 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
116 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
117 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
128 FIO_RAND_FILE_SIZE_OFF,
131 FIO_RAND_SEQ_RAND_READ_OFF,
132 FIO_RAND_SEQ_RAND_WRITE_OFF,
133 FIO_RAND_SEQ_RAND_TRIM_OFF,
134 FIO_RAND_START_DELAY,
136 FIO_RAND_POISSON_OFF,
138 FIO_RAND_POISSON2_OFF,
139 FIO_RAND_POISSON3_OFF,
147 RATE_PROCESS_LINEAR = 0,
148 RATE_PROCESS_POISSON = 1,
159 * Per-thread/process specific data. Only used for the network client
162 void sk_out_assign(struct sk_out *);
163 void sk_out_drop(void);
165 struct zone_split_index {
167 uint8_t size_perc_prev;
172 #define FIO_MAX_OPEN_ZBD_ZONES 128
175 * This describes a single thread/process executing a fio job.
178 struct flist_head opt_list;
180 struct thread_options o;
183 unsigned int thread_number;
184 unsigned int subjob_number;
185 unsigned int groupid;
186 struct thread_stat ts __attribute__ ((aligned(8)));
190 struct io_log *slat_log;
191 struct io_log *clat_log;
192 struct io_log *clat_hist_log;
193 struct io_log *lat_log;
194 struct io_log *bw_log;
195 struct io_log *iops_log;
197 struct workqueue log_compress_wq;
199 struct thread_data *parent;
201 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
202 struct timespec bw_sample_time;
204 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
205 struct timespec iops_sample_time;
207 volatile int update_rusage;
208 struct fio_sem *rusage_sem;
209 struct rusage ru_start;
210 struct rusage ru_end;
212 struct fio_file **files;
213 unsigned char *file_locks;
214 unsigned int files_size;
215 unsigned int files_index;
216 unsigned int nr_open_files;
217 unsigned int nr_done_files;
219 unsigned int next_file;
220 struct frand_state next_file_state;
223 struct zipf_state next_file_zipf;
224 struct gauss_state next_file_gauss;
237 size_t orig_buffer_size;
238 volatile int runstate;
239 volatile bool terminate;
241 enum fio_ddir last_ddir;
248 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
250 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
251 struct frand_state verify_state;
252 struct frand_state trim_state;
253 struct frand_state delay_state;
255 struct frand_state buf_state;
256 struct frand_state buf_state_prev;
257 struct frand_state dedupe_state;
258 struct frand_state zone_state;
260 struct zone_split_index **zone_state_index;
262 unsigned int verify_batch;
263 unsigned int trim_batch;
265 struct thread_io_list *vstate;
270 * IO engine hooks, contains everything needed to submit an io_u
271 * to any of the available IO engines.
273 struct ioengine_ops *io_ops;
277 * IO engine private data and dlhandle.
280 void *io_ops_dlhandle;
283 * Queue depth of io_u's that fio MIGHT do
285 unsigned int cur_depth;
288 * io_u's about to be committed
290 unsigned int io_u_queued;
293 * io_u's submitted but not completed yet
295 unsigned int io_u_in_flight;
298 * List of free and busy io_u's
300 struct io_u_ring io_u_requeues;
301 struct io_u_queue io_u_freelist;
302 struct io_u_queue io_u_all;
303 pthread_mutex_t io_u_lock;
304 pthread_cond_t free_cond;
307 * async verify offload
309 struct flist_head verify_list;
310 pthread_t *verify_threads;
311 unsigned int nr_verify_threads;
312 pthread_cond_t verify_cond;
313 int verify_thread_exit;
318 uint64_t rate_bps[DDIR_RWDIR_CNT];
319 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
320 unsigned long rate_bytes[DDIR_RWDIR_CNT];
321 unsigned long rate_blocks[DDIR_RWDIR_CNT];
322 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
323 struct timespec lastrate[DDIR_RWDIR_CNT];
324 int64_t last_usec[DDIR_RWDIR_CNT];
325 struct frand_state poisson_state[DDIR_RWDIR_CNT];
328 * Enforced rate submission/completion workqueue
330 struct workqueue io_wq;
332 uint64_t total_io_size;
333 uint64_t fill_device_size;
338 uint64_t io_issues[DDIR_RWDIR_CNT];
339 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
345 uint64_t io_blocks[DDIR_RWDIR_CNT];
346 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
347 uint64_t io_bytes[DDIR_RWDIR_CNT];
348 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
349 uint64_t io_skip_bytes;
352 uint64_t bytes_done[DDIR_RWDIR_CNT];
355 * State for random io, a bitmap of blocks done vs not done
357 struct frand_state random_state;
359 struct timespec start; /* start of this loop */
360 struct timespec epoch; /* time job was started */
361 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
362 struct timespec last_issue;
364 struct timespec ts_cache;
365 struct timespec terminate_time;
366 unsigned int ts_cache_nr;
367 unsigned int ts_cache_mask;
371 * Time since last latency_window was started
373 struct timespec latency_ts;
374 unsigned int latency_qd;
375 unsigned int latency_qd_high;
376 unsigned int latency_qd_low;
377 unsigned int latency_failed;
378 uint64_t latency_ios;
382 * read/write mixed workload state
384 struct frand_state rwmix_state;
385 unsigned long rwmix_issues;
386 enum fio_ddir rwmix_ddir;
387 unsigned int ddir_seq_nr;
390 * rand/seq mixed workload state
392 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
395 * IO history logs for verification. We use a tree for sorting,
396 * if we are overwriting. Otherwise just use a fifo.
398 struct rb_root io_hist_tree;
399 struct flist_head io_hist_list;
400 unsigned long io_hist_len;
405 struct flist_head io_log_list;
407 unsigned int io_log_current;
408 unsigned int io_log_checkmark;
409 unsigned int io_log_highmark;
410 struct timespec io_log_highmark_time;
413 * For tracking/handling discards
415 struct flist_head trim_list;
416 unsigned long trim_entries;
419 * for fileservice, how often to switch to a new file
421 unsigned int file_service_nr;
422 unsigned int file_service_left;
423 struct fio_file *file_service_file;
425 unsigned int sync_file_range_nr;
428 * For generating file sizes
430 struct frand_state file_size_state;
435 unsigned int total_err_count;
438 struct fio_flow *flow;
441 * Can be overloaded by profiles
443 struct prof_io_ops prof_io_ops;
448 struct steadystate_data ss;
450 char verror[FIO_VERROR_SIZE];
454 * for GPU memory management
460 CUdeviceptr dev_mem_ptr;
466 * when should interactive ETA output be generated
474 #define __td_verror(td, err, msg, func) \
476 unsigned int ____e = (err); \
479 (td)->error = ____e; \
480 if (!(td)->first_error) \
481 nowarn_snprintf(td->verror, sizeof(td->verror), \
482 "file:%s:%d, func=%s, error=%s", \
483 __FILE__, __LINE__, (func), (msg)); \
487 #define td_clear_error(td) do { \
490 (td)->parent->error = 0; \
493 #define td_verror(td, err, func) do { \
494 __td_verror((td), (err), strerror((err)), (func)); \
496 __td_verror((td)->parent, (err), strerror((err)), (func)); \
499 #define td_vmsg(td, err, msg, func) do { \
500 __td_verror((td), (err), (msg), (func)); \
502 __td_verror((td)->parent, (err), (msg), (func)); \
505 #define __fio_stringify_1(x) #x
506 #define __fio_stringify(x) __fio_stringify_1(x)
508 extern bool exitall_on_terminate;
509 extern unsigned int thread_number;
510 extern unsigned int stat_number;
513 extern int output_format;
514 extern int append_terse_output;
515 extern int temp_stall_ts;
516 extern uintptr_t page_mask, page_size;
517 extern bool read_only;
518 extern int eta_print;
519 extern int eta_new_line;
520 extern unsigned int eta_interval_msec;
521 extern unsigned long done_secs;
522 extern int fio_gtod_offload;
523 extern int fio_gtod_cpu;
524 extern enum fio_cs fio_clock_source;
525 extern int fio_clock_source_set;
526 extern int warnings_fatal;
527 extern int terse_version;
528 extern bool is_backend;
529 extern bool is_local_backend;
530 extern int nr_clients;
531 extern bool log_syslog;
532 extern int status_interval;
533 extern const char fio_version_string[];
534 extern char *trigger_file;
535 extern char *trigger_cmd;
536 extern char *trigger_remote_cmd;
537 extern long long trigger_timeout;
538 extern char *aux_path;
540 extern struct thread_data *threads;
542 static inline bool is_running_backend(void)
544 return is_backend || is_local_backend;
547 extern bool eta_time_within_slack(unsigned int time);
549 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
551 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)) &&
552 !(io_u->ddir == DDIR_TRIM && !td_trim(td)));
555 #define REAL_MAX_JOBS 4096
557 static inline bool should_fsync(struct thread_data *td)
559 if (td->last_was_sync)
561 if (td_write(td) || td->o.override_sync)
568 * Init/option functions
570 extern int __must_check fio_init_options(void);
571 extern int __must_check parse_options(int, char **);
572 extern int parse_jobs_ini(char *, int, int, int);
573 extern int parse_cmd_line(int, char **, int);
574 extern int fio_backend(struct sk_out *);
575 extern void reset_fio_state(void);
576 extern void clear_io_state(struct thread_data *, int);
577 extern int fio_options_parse(struct thread_data *, char **, int);
578 extern void fio_keywords_init(void);
579 extern void fio_keywords_exit(void);
580 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
581 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
582 extern void fio_fill_default_options(struct thread_data *);
583 extern int fio_show_option_help(const char *);
584 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
585 extern void fio_options_dup_and_init(struct option *);
586 extern char *fio_option_dup_subs(const char *);
587 extern void fio_options_mem_dupe(struct thread_data *);
588 extern void td_fill_rand_seeds(struct thread_data *);
589 extern void td_fill_verify_state_seed(struct thread_data *);
590 extern void add_job_opts(const char **, int);
591 extern int ioengine_load(struct thread_data *);
592 extern bool parse_dryrun(void);
593 extern int fio_running_or_pending_io_threads(void);
594 extern int fio_set_fd_nonblocking(int, const char *);
595 extern void sig_show_status(int sig);
596 extern struct thread_data *get_global_options(void);
598 extern uintptr_t page_mask;
599 extern uintptr_t page_size;
600 extern int initialize_fio(char *envp[]);
601 extern void deinitialize_fio(void);
603 #define FIO_GETOPT_JOB 0x89000000
604 #define FIO_GETOPT_IOENGINE 0x98000000
605 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
610 extern void print_thread_status(void);
611 extern void print_status_init(int);
612 extern char *fio_uint_to_kmg(unsigned int val);
615 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
616 * will never back again. It may cycle between running/verififying/fsyncing.
617 * Once the thread reaches TD_EXITED, it is just waiting for the core to
637 #define TD_ENG_FLAG_SHIFT 17
638 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
640 static inline void td_set_ioengine_flags(struct thread_data *td)
642 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
643 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
646 static inline bool td_ioengine_flagged(struct thread_data *td,
647 enum fio_ioengine_flags flags)
649 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
652 extern void td_set_runstate(struct thread_data *, int);
653 extern int td_bump_runstate(struct thread_data *, int);
654 extern void td_restore_runstate(struct thread_data *, int);
655 extern const char *runstate_to_name(int runstate);
658 * Allow 60 seconds for a job to quit on its own, otherwise reap with
661 #define FIO_REAP_TIMEOUT 300
663 #define TERMINATE_ALL (-1U)
664 extern void fio_terminate_threads(unsigned int);
665 extern void fio_mark_td_terminate(struct thread_data *);
670 extern int __must_check fio_pin_memory(struct thread_data *);
671 extern void fio_unpin_memory(struct thread_data *);
672 extern int __must_check allocate_io_mem(struct thread_data *);
673 extern void free_io_mem(struct thread_data *);
674 extern void free_threads_shm(void);
677 #define PTR_ALIGN(ptr, mask) \
678 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
682 * Reset stats after ramp time completes
684 extern void reset_all_stats(struct thread_data *);
686 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
687 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
688 struct timespec *comp_time);
691 * Latency target helpers
693 extern void lat_target_check(struct thread_data *);
694 extern void lat_target_init(struct thread_data *);
695 extern void lat_target_reset(struct thread_data *);
698 * Iterates all threads/processes within all the defined jobs
700 #define for_each_td(td, i) \
701 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
702 #define for_each_file(td, f, i) \
703 if ((td)->files_index) \
704 for ((i) = 0, (f) = (td)->files[0]; \
705 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
708 #define fio_assert(td, cond) do { \
711 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
712 td_set_runstate((td), TD_EXITED); \
713 (td)->error = EFAULT; \
718 static inline bool fio_fill_issue_time(struct thread_data *td)
720 if (td->o.read_iolog_file ||
721 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
727 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
729 struct thread_options *o = &td->o;
732 * If some rate setting was given, we need to check it
734 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
735 o->rate_iops_min[ddir])
741 static inline bool __should_check_rate(struct thread_data *td)
743 return (td->flags & TD_F_CHECK_RATE) != 0;
746 static inline bool should_check_rate(struct thread_data *td)
748 if (!__should_check_rate(td))
751 return ddir_rw_sum(td->bytes_done) != 0;
754 static inline unsigned long long td_max_bs(struct thread_data *td)
756 unsigned long long max_bs;
758 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
759 return max(td->o.max_bs[DDIR_TRIM], max_bs);
762 static inline unsigned long long td_min_bs(struct thread_data *td)
764 unsigned long long min_bs;
766 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
767 return min(td->o.min_bs[DDIR_TRIM], min_bs);
770 static inline bool td_async_processing(struct thread_data *td)
772 return (td->flags & TD_F_NEED_LOCK) != 0;
776 * We currently only need to do locking if we have verifier threads
777 * accessing our internal structures too
779 static inline void __td_io_u_lock(struct thread_data *td)
781 pthread_mutex_lock(&td->io_u_lock);
784 static inline void __td_io_u_unlock(struct thread_data *td)
786 pthread_mutex_unlock(&td->io_u_lock);
789 static inline void td_io_u_free_notify(struct thread_data *td)
791 if (td_async_processing(td))
792 pthread_cond_signal(&td->free_cond);
795 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
798 if (!td_async_processing(td))
801 __sync_fetch_and_and(flags, ~value);
804 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
807 if (!td_async_processing(td))
810 __sync_fetch_and_or(flags, value);
813 extern const char *fio_get_arch_string(int);
814 extern const char *fio_get_os_string(int);
817 __FIO_OUTPUT_TERSE = 0,
818 __FIO_OUTPUT_JSON = 1,
819 __FIO_OUTPUT_NORMAL = 2,
820 __FIO_OUTPUT_JSON_PLUS = 3,
823 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
824 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
825 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
826 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
830 FIO_RAND_DIST_RANDOM = 0,
832 FIO_RAND_DIST_PARETO,
835 FIO_RAND_DIST_ZONED_ABS,
838 #define FIO_DEF_ZIPF 1.1
839 #define FIO_DEF_PARETO 0.2
842 FIO_RAND_GEN_TAUSWORTHE = 0,
844 FIO_RAND_GEN_TAUSWORTHE64,
852 extern void exec_trigger(const char *);
853 extern void check_trigger_file(void);
855 extern bool in_flight_overlap(struct io_u_queue *q, struct io_u *io_u);
856 extern pthread_mutex_t overlap_check;