8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
28 #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"
48 #ifdef CONFIG_SOLARISAIO
49 #include <sys/asynch.h>
53 #include <linux/mempolicy.h>
57 * "local" is pseudo-policy
59 #define MPOL_LOCAL MPOL_MAX
67 * offset generator types
75 __TD_F_VER_BACKLOG = 0,
78 __TD_F_REFILL_BUFFERS,
79 __TD_F_SCRAMBLE_BUFFERS,
93 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
94 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
95 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
96 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
97 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
98 TD_F_VER_NONE = 1U << __TD_F_VER_NONE,
99 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
100 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
101 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
102 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
103 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
104 TD_F_CHILD = 1U << __TD_F_CHILD,
105 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
106 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
107 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
118 FIO_RAND_FILE_SIZE_OFF,
121 FIO_RAND_SEQ_RAND_READ_OFF,
122 FIO_RAND_SEQ_RAND_WRITE_OFF,
123 FIO_RAND_SEQ_RAND_TRIM_OFF,
124 FIO_RAND_START_DELAY,
126 FIO_RAND_POISSON_OFF,
128 FIO_RAND_POISSON2_OFF,
129 FIO_RAND_POISSON3_OFF,
137 RATE_PROCESS_LINEAR = 0,
138 RATE_PROCESS_POISSON = 1,
149 * Per-thread/process specific data. Only used for the network client
152 void sk_out_assign(struct sk_out *);
153 void sk_out_drop(void);
155 struct zone_split_index {
157 uint8_t size_perc_prev;
161 * This describes a single thread/process executing a fio job.
164 struct flist_head opt_list;
166 struct thread_options o;
169 unsigned int thread_number;
170 unsigned int subjob_number;
171 unsigned int groupid;
172 struct thread_stat ts __attribute__ ((aligned(8)));
176 struct io_log *slat_log;
177 struct io_log *clat_log;
178 struct io_log *clat_hist_log;
179 struct io_log *lat_log;
180 struct io_log *bw_log;
181 struct io_log *iops_log;
183 struct workqueue log_compress_wq;
185 struct thread_data *parent;
187 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
188 struct timespec bw_sample_time;
190 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
191 struct timespec iops_sample_time;
193 volatile int update_rusage;
194 struct fio_mutex *rusage_sem;
195 struct rusage ru_start;
196 struct rusage ru_end;
198 struct fio_file **files;
199 unsigned char *file_locks;
200 unsigned int files_size;
201 unsigned int files_index;
202 unsigned int nr_open_files;
203 unsigned int nr_done_files;
204 unsigned int nr_normal_files;
206 unsigned int next_file;
207 struct frand_state next_file_state;
210 struct zipf_state next_file_zipf;
211 struct gauss_state next_file_gauss;
224 size_t orig_buffer_size;
225 volatile int terminate;
226 volatile int runstate;
227 unsigned int last_was_sync;
228 enum fio_ddir last_ddir;
235 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
237 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
238 struct frand_state verify_state;
239 struct frand_state trim_state;
240 struct frand_state delay_state;
242 struct frand_state buf_state;
243 struct frand_state buf_state_prev;
244 struct frand_state dedupe_state;
245 struct frand_state zone_state;
247 struct zone_split_index **zone_state_index;
249 unsigned int verify_batch;
250 unsigned int trim_batch;
252 struct thread_io_list *vstate;
257 * IO engine hooks, contains everything needed to submit an io_u
258 * to any of the available IO engines.
260 struct ioengine_ops *io_ops;
264 * IO engine private data and dlhandle.
267 void *io_ops_dlhandle;
270 * Queue depth of io_u's that fio MIGHT do
272 unsigned int cur_depth;
275 * io_u's about to be committed
277 unsigned int io_u_queued;
280 * io_u's submitted but not completed yet
282 unsigned int io_u_in_flight;
285 * List of free and busy io_u's
287 struct io_u_ring io_u_requeues;
288 struct io_u_queue io_u_freelist;
289 struct io_u_queue io_u_all;
290 pthread_mutex_t io_u_lock;
291 pthread_cond_t free_cond;
294 * async verify offload
296 struct flist_head verify_list;
297 pthread_t *verify_threads;
298 unsigned int nr_verify_threads;
299 pthread_cond_t verify_cond;
300 int verify_thread_exit;
305 uint64_t rate_bps[DDIR_RWDIR_CNT];
306 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
307 unsigned long rate_bytes[DDIR_RWDIR_CNT];
308 unsigned long rate_blocks[DDIR_RWDIR_CNT];
309 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
310 struct timespec lastrate[DDIR_RWDIR_CNT];
311 int64_t last_usec[DDIR_RWDIR_CNT];
312 struct frand_state poisson_state[DDIR_RWDIR_CNT];
315 * Enforced rate submission/completion workqueue
317 struct workqueue io_wq;
319 uint64_t total_io_size;
320 uint64_t fill_device_size;
325 uint64_t io_issues[DDIR_RWDIR_CNT];
326 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
332 uint64_t io_blocks[DDIR_RWDIR_CNT];
333 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
334 uint64_t io_bytes[DDIR_RWDIR_CNT];
335 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
336 uint64_t io_skip_bytes;
338 struct fio_mutex *mutex;
339 uint64_t bytes_done[DDIR_RWDIR_CNT];
342 * State for random io, a bitmap of blocks done vs not done
344 struct frand_state random_state;
346 struct timespec start; /* start of this loop */
347 struct timespec epoch; /* time job was started */
348 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
349 struct timespec last_issue;
351 struct timespec ts_cache;
352 struct timespec terminate_time;
353 unsigned int ts_cache_nr;
354 unsigned int ts_cache_mask;
358 * Time since last latency_window was started
360 struct timespec latency_ts;
361 unsigned int latency_qd;
362 unsigned int latency_qd_high;
363 unsigned int latency_qd_low;
364 unsigned int latency_failed;
365 uint64_t latency_ios;
369 * read/write mixed workload state
371 struct frand_state rwmix_state;
372 unsigned long rwmix_issues;
373 enum fio_ddir rwmix_ddir;
374 unsigned int ddir_seq_nr;
377 * rand/seq mixed workload state
379 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
382 * IO history logs for verification. We use a tree for sorting,
383 * if we are overwriting. Otherwise just use a fifo.
385 struct rb_root io_hist_tree;
386 struct flist_head io_hist_list;
387 unsigned long io_hist_len;
392 struct flist_head io_log_list;
395 * For tracking/handling discards
397 struct flist_head trim_list;
398 unsigned long trim_entries;
400 struct flist_head next_rand_list;
403 * for fileservice, how often to switch to a new file
405 unsigned int file_service_nr;
406 unsigned int file_service_left;
407 struct fio_file *file_service_file;
409 unsigned int sync_file_range_nr;
412 * For generating file sizes
414 struct frand_state file_size_state;
419 unsigned int total_err_count;
422 struct fio_flow *flow;
425 * Can be overloaded by profiles
427 struct prof_io_ops prof_io_ops;
432 struct steadystate_data ss;
434 char verror[FIO_VERROR_SIZE];
438 * for GPU memory management
444 CUdeviceptr dev_mem_ptr;
450 * when should interactive ETA output be generated
458 #define __td_verror(td, err, msg, func) \
460 unsigned int ____e = (err); \
463 (td)->error = ____e; \
464 if (!(td)->first_error) \
465 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
469 #define td_clear_error(td) do { \
472 (td)->parent->error = 0; \
475 #define td_verror(td, err, func) do { \
476 __td_verror((td), (err), strerror((err)), (func)); \
478 __td_verror((td)->parent, (err), strerror((err)), (func)); \
481 #define td_vmsg(td, err, msg, func) do { \
482 __td_verror((td), (err), (msg), (func)); \
484 __td_verror((td)->parent, (err), (msg), (func)); \
487 #define __fio_stringify_1(x) #x
488 #define __fio_stringify(x) __fio_stringify_1(x)
490 extern int exitall_on_terminate;
491 extern unsigned int thread_number;
492 extern unsigned int stat_number;
495 extern int output_format;
496 extern int append_terse_output;
497 extern int temp_stall_ts;
498 extern uintptr_t page_mask, page_size;
499 extern int read_only;
500 extern int eta_print;
501 extern int eta_new_line;
502 extern unsigned long done_secs;
503 extern int fio_gtod_offload;
504 extern int fio_gtod_cpu;
505 extern enum fio_cs fio_clock_source;
506 extern int fio_clock_source_set;
507 extern int warnings_fatal;
508 extern int terse_version;
509 extern int is_backend;
510 extern int nr_clients;
511 extern int log_syslog;
512 extern int status_interval;
513 extern const char fio_version_string[];
514 extern char *trigger_file;
515 extern char *trigger_cmd;
516 extern char *trigger_remote_cmd;
517 extern long long trigger_timeout;
518 extern char *aux_path;
520 extern struct thread_data *threads;
522 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
524 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
527 #define REAL_MAX_JOBS 4096
529 static inline int should_fsync(struct thread_data *td)
531 if (td->last_was_sync)
533 if (td_write(td) || td->o.override_sync)
540 * Init/option functions
542 extern int __must_check fio_init_options(void);
543 extern int __must_check parse_options(int, char **);
544 extern int parse_jobs_ini(char *, int, int, int);
545 extern int parse_cmd_line(int, char **, int);
546 extern int fio_backend(struct sk_out *);
547 extern void reset_fio_state(void);
548 extern void clear_io_state(struct thread_data *, int);
549 extern int fio_options_parse(struct thread_data *, char **, int);
550 extern void fio_keywords_init(void);
551 extern void fio_keywords_exit(void);
552 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
553 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
554 extern void fio_fill_default_options(struct thread_data *);
555 extern int fio_show_option_help(const char *);
556 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
557 extern void fio_options_dup_and_init(struct option *);
558 extern void fio_options_mem_dupe(struct thread_data *);
559 extern void td_fill_rand_seeds(struct thread_data *);
560 extern void td_fill_verify_state_seed(struct thread_data *);
561 extern void add_job_opts(const char **, int);
562 extern int ioengine_load(struct thread_data *);
563 extern bool parse_dryrun(void);
564 extern int fio_running_or_pending_io_threads(void);
565 extern int fio_set_fd_nonblocking(int, const char *);
566 extern void sig_show_status(int sig);
567 extern struct thread_data *get_global_options(void);
569 extern uintptr_t page_mask;
570 extern uintptr_t page_size;
571 extern int initialize_fio(char *envp[]);
572 extern void deinitialize_fio(void);
574 #define FIO_GETOPT_JOB 0x89000000
575 #define FIO_GETOPT_IOENGINE 0x98000000
576 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
581 extern void print_thread_status(void);
582 extern void print_status_init(int);
583 extern char *fio_uint_to_kmg(unsigned int val);
586 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
587 * will never back again. It may cycle between running/verififying/fsyncing.
588 * Once the thread reaches TD_EXITED, it is just waiting for the core to
608 #define TD_ENG_FLAG_SHIFT 16
609 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
611 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
613 return (enum fio_ioengine_flags)
614 ((td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK);
617 static inline void td_set_ioengine_flags(struct thread_data *td)
619 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
620 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
623 static inline bool td_ioengine_flagged(struct thread_data *td,
624 enum fio_ioengine_flags flags)
626 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
629 extern void td_set_runstate(struct thread_data *, int);
630 extern int td_bump_runstate(struct thread_data *, int);
631 extern void td_restore_runstate(struct thread_data *, int);
632 extern const char *runstate_to_name(int runstate);
635 * Allow 60 seconds for a job to quit on its own, otherwise reap with
638 #define FIO_REAP_TIMEOUT 300
640 #define TERMINATE_ALL (-1U)
641 extern void fio_terminate_threads(unsigned int);
642 extern void fio_mark_td_terminate(struct thread_data *);
647 extern int __must_check fio_pin_memory(struct thread_data *);
648 extern void fio_unpin_memory(struct thread_data *);
649 extern int __must_check allocate_io_mem(struct thread_data *);
650 extern void free_io_mem(struct thread_data *);
651 extern void free_threads_shm(void);
654 #define PTR_ALIGN(ptr, mask) \
655 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
659 * Reset stats after ramp time completes
661 extern void reset_all_stats(struct thread_data *);
663 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
664 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
665 struct timespec *comp_time);
668 * Latency target helpers
670 extern void lat_target_check(struct thread_data *);
671 extern void lat_target_init(struct thread_data *);
672 extern void lat_target_reset(struct thread_data *);
675 * Iterates all threads/processes within all the defined jobs
677 #define for_each_td(td, i) \
678 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
679 #define for_each_file(td, f, i) \
680 if ((td)->files_index) \
681 for ((i) = 0, (f) = (td)->files[0]; \
682 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
685 #define fio_assert(td, cond) do { \
688 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
689 td_set_runstate((td), TD_EXITED); \
690 (td)->error = EFAULT; \
695 static inline bool fio_fill_issue_time(struct thread_data *td)
697 if (td->o.read_iolog_file ||
698 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
704 static inline bool __should_check_rate(struct thread_data *td,
707 struct thread_options *o = &td->o;
710 * If some rate setting was given, we need to check it
712 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
713 o->rate_iops_min[ddir])
719 static inline bool should_check_rate(struct thread_data *td)
721 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
723 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
725 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
731 static inline unsigned int td_max_bs(struct thread_data *td)
735 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
736 return max(td->o.max_bs[DDIR_TRIM], max_bs);
739 static inline unsigned int td_min_bs(struct thread_data *td)
743 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
744 return min(td->o.min_bs[DDIR_TRIM], min_bs);
747 static inline bool td_async_processing(struct thread_data *td)
749 return (td->flags & TD_F_NEED_LOCK) != 0;
753 * We currently only need to do locking if we have verifier threads
754 * accessing our internal structures too
756 static inline void td_io_u_lock(struct thread_data *td)
758 if (td_async_processing(td))
759 pthread_mutex_lock(&td->io_u_lock);
762 static inline void td_io_u_unlock(struct thread_data *td)
764 if (td_async_processing(td))
765 pthread_mutex_unlock(&td->io_u_lock);
768 static inline void td_io_u_free_notify(struct thread_data *td)
770 if (td_async_processing(td))
771 pthread_cond_signal(&td->free_cond);
774 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
777 if (!td_async_processing(td))
780 __sync_fetch_and_and(flags, ~value);
783 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
786 if (!td_async_processing(td))
789 __sync_fetch_and_or(flags, value);
792 extern const char *fio_get_arch_string(int);
793 extern const char *fio_get_os_string(int);
796 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
797 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
801 __FIO_OUTPUT_TERSE = 0,
802 __FIO_OUTPUT_JSON = 1,
803 __FIO_OUTPUT_NORMAL = 2,
804 __FIO_OUTPUT_JSON_PLUS = 3,
807 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
808 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
809 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
810 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
814 FIO_RAND_DIST_RANDOM = 0,
816 FIO_RAND_DIST_PARETO,
821 #define FIO_DEF_ZIPF 1.1
822 #define FIO_DEF_PARETO 0.2
825 FIO_RAND_GEN_TAUSWORTHE = 0,
827 FIO_RAND_GEN_TAUSWORTHE64,
835 extern void exec_trigger(const char *);
836 extern void check_trigger_file(void);