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,
92 __TD_F_LAST, /* not a real bit, keep last */
96 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
97 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
98 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
99 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
100 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
101 TD_F_VER_NONE = 1U << __TD_F_VER_NONE,
102 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
103 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
104 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
105 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
106 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
107 TD_F_CHILD = 1U << __TD_F_CHILD,
108 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
109 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
110 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
111 TD_F_DIRS_CREATED = 1U << __TD_F_DIRS_CREATED,
112 TD_F_CHECK_RATE = 1U << __TD_F_CHECK_RATE,
123 FIO_RAND_FILE_SIZE_OFF,
126 FIO_RAND_SEQ_RAND_READ_OFF,
127 FIO_RAND_SEQ_RAND_WRITE_OFF,
128 FIO_RAND_SEQ_RAND_TRIM_OFF,
129 FIO_RAND_START_DELAY,
131 FIO_RAND_POISSON_OFF,
133 FIO_RAND_POISSON2_OFF,
134 FIO_RAND_POISSON3_OFF,
142 RATE_PROCESS_LINEAR = 0,
143 RATE_PROCESS_POISSON = 1,
154 * Per-thread/process specific data. Only used for the network client
157 void sk_out_assign(struct sk_out *);
158 void sk_out_drop(void);
160 struct zone_split_index {
162 uint8_t size_perc_prev;
168 * This describes a single thread/process executing a fio job.
171 struct flist_head opt_list;
173 struct thread_options o;
176 unsigned int thread_number;
177 unsigned int subjob_number;
178 unsigned int groupid;
179 struct thread_stat ts __attribute__ ((aligned(8)));
183 struct io_log *slat_log;
184 struct io_log *clat_log;
185 struct io_log *clat_hist_log;
186 struct io_log *lat_log;
187 struct io_log *bw_log;
188 struct io_log *iops_log;
190 struct workqueue log_compress_wq;
192 struct thread_data *parent;
194 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
195 struct timespec bw_sample_time;
197 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
198 struct timespec iops_sample_time;
200 volatile int update_rusage;
201 struct fio_mutex *rusage_sem;
202 struct rusage ru_start;
203 struct rusage ru_end;
205 struct fio_file **files;
206 unsigned char *file_locks;
207 unsigned int files_size;
208 unsigned int files_index;
209 unsigned int nr_open_files;
210 unsigned int nr_done_files;
211 unsigned int nr_normal_files;
213 unsigned int next_file;
214 struct frand_state next_file_state;
217 struct zipf_state next_file_zipf;
218 struct gauss_state next_file_gauss;
231 size_t orig_buffer_size;
232 volatile int terminate;
233 volatile int runstate;
234 unsigned int last_was_sync;
235 enum fio_ddir last_ddir;
242 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
244 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
245 struct frand_state verify_state;
246 struct frand_state trim_state;
247 struct frand_state delay_state;
249 struct frand_state buf_state;
250 struct frand_state buf_state_prev;
251 struct frand_state dedupe_state;
252 struct frand_state zone_state;
254 struct zone_split_index **zone_state_index;
256 unsigned int verify_batch;
257 unsigned int trim_batch;
259 struct thread_io_list *vstate;
264 * IO engine hooks, contains everything needed to submit an io_u
265 * to any of the available IO engines.
267 struct ioengine_ops *io_ops;
271 * IO engine private data and dlhandle.
274 void *io_ops_dlhandle;
277 * Queue depth of io_u's that fio MIGHT do
279 unsigned int cur_depth;
282 * io_u's about to be committed
284 unsigned int io_u_queued;
287 * io_u's submitted but not completed yet
289 unsigned int io_u_in_flight;
292 * List of free and busy io_u's
294 struct io_u_ring io_u_requeues;
295 struct io_u_queue io_u_freelist;
296 struct io_u_queue io_u_all;
297 pthread_mutex_t io_u_lock;
298 pthread_cond_t free_cond;
301 * async verify offload
303 struct flist_head verify_list;
304 pthread_t *verify_threads;
305 unsigned int nr_verify_threads;
306 pthread_cond_t verify_cond;
307 int verify_thread_exit;
312 uint64_t rate_bps[DDIR_RWDIR_CNT];
313 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
314 unsigned long rate_bytes[DDIR_RWDIR_CNT];
315 unsigned long rate_blocks[DDIR_RWDIR_CNT];
316 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
317 struct timespec lastrate[DDIR_RWDIR_CNT];
318 int64_t last_usec[DDIR_RWDIR_CNT];
319 struct frand_state poisson_state[DDIR_RWDIR_CNT];
322 * Enforced rate submission/completion workqueue
324 struct workqueue io_wq;
326 uint64_t total_io_size;
327 uint64_t fill_device_size;
332 uint64_t io_issues[DDIR_RWDIR_CNT];
333 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
339 uint64_t io_blocks[DDIR_RWDIR_CNT];
340 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
341 uint64_t io_bytes[DDIR_RWDIR_CNT];
342 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
343 uint64_t io_skip_bytes;
345 struct fio_mutex *mutex;
346 uint64_t bytes_done[DDIR_RWDIR_CNT];
349 * State for random io, a bitmap of blocks done vs not done
351 struct frand_state random_state;
353 struct timespec start; /* start of this loop */
354 struct timespec epoch; /* time job was started */
355 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
356 struct timespec last_issue;
358 struct timespec ts_cache;
359 struct timespec terminate_time;
360 unsigned int ts_cache_nr;
361 unsigned int ts_cache_mask;
365 * Time since last latency_window was started
367 struct timespec latency_ts;
368 unsigned int latency_qd;
369 unsigned int latency_qd_high;
370 unsigned int latency_qd_low;
371 unsigned int latency_failed;
372 uint64_t latency_ios;
376 * read/write mixed workload state
378 struct frand_state rwmix_state;
379 unsigned long rwmix_issues;
380 enum fio_ddir rwmix_ddir;
381 unsigned int ddir_seq_nr;
384 * rand/seq mixed workload state
386 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
389 * IO history logs for verification. We use a tree for sorting,
390 * if we are overwriting. Otherwise just use a fifo.
392 struct rb_root io_hist_tree;
393 struct flist_head io_hist_list;
394 unsigned long io_hist_len;
399 struct flist_head io_log_list;
402 * For tracking/handling discards
404 struct flist_head trim_list;
405 unsigned long trim_entries;
407 struct flist_head next_rand_list;
410 * for fileservice, how often to switch to a new file
412 unsigned int file_service_nr;
413 unsigned int file_service_left;
414 struct fio_file *file_service_file;
416 unsigned int sync_file_range_nr;
419 * For generating file sizes
421 struct frand_state file_size_state;
426 unsigned int total_err_count;
429 struct fio_flow *flow;
432 * Can be overloaded by profiles
434 struct prof_io_ops prof_io_ops;
439 struct steadystate_data ss;
441 char verror[FIO_VERROR_SIZE];
445 * for GPU memory management
451 CUdeviceptr dev_mem_ptr;
457 * when should interactive ETA output be generated
465 #define __td_verror(td, err, msg, func) \
467 unsigned int ____e = (err); \
470 (td)->error = ____e; \
471 if (!(td)->first_error) \
472 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
476 #define td_clear_error(td) do { \
479 (td)->parent->error = 0; \
482 #define td_verror(td, err, func) do { \
483 __td_verror((td), (err), strerror((err)), (func)); \
485 __td_verror((td)->parent, (err), strerror((err)), (func)); \
488 #define td_vmsg(td, err, msg, func) do { \
489 __td_verror((td), (err), (msg), (func)); \
491 __td_verror((td)->parent, (err), (msg), (func)); \
494 #define __fio_stringify_1(x) #x
495 #define __fio_stringify(x) __fio_stringify_1(x)
497 extern int exitall_on_terminate;
498 extern unsigned int thread_number;
499 extern unsigned int stat_number;
502 extern int output_format;
503 extern int append_terse_output;
504 extern int temp_stall_ts;
505 extern uintptr_t page_mask, page_size;
506 extern int read_only;
507 extern int eta_print;
508 extern int eta_new_line;
509 extern unsigned long done_secs;
510 extern int fio_gtod_offload;
511 extern int fio_gtod_cpu;
512 extern enum fio_cs fio_clock_source;
513 extern int fio_clock_source_set;
514 extern int warnings_fatal;
515 extern int terse_version;
516 extern int is_backend;
517 extern int nr_clients;
518 extern int log_syslog;
519 extern int status_interval;
520 extern const char fio_version_string[];
521 extern char *trigger_file;
522 extern char *trigger_cmd;
523 extern char *trigger_remote_cmd;
524 extern long long trigger_timeout;
525 extern char *aux_path;
527 extern struct thread_data *threads;
529 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
531 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
534 #define REAL_MAX_JOBS 4096
536 static inline int should_fsync(struct thread_data *td)
538 if (td->last_was_sync)
540 if (td_write(td) || td->o.override_sync)
547 * Init/option functions
549 extern int __must_check fio_init_options(void);
550 extern int __must_check parse_options(int, char **);
551 extern int parse_jobs_ini(char *, int, int, int);
552 extern int parse_cmd_line(int, char **, int);
553 extern int fio_backend(struct sk_out *);
554 extern void reset_fio_state(void);
555 extern void clear_io_state(struct thread_data *, int);
556 extern int fio_options_parse(struct thread_data *, char **, int);
557 extern void fio_keywords_init(void);
558 extern void fio_keywords_exit(void);
559 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
560 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
561 extern void fio_fill_default_options(struct thread_data *);
562 extern int fio_show_option_help(const char *);
563 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
564 extern void fio_options_dup_and_init(struct option *);
565 extern void fio_options_mem_dupe(struct thread_data *);
566 extern void td_fill_rand_seeds(struct thread_data *);
567 extern void td_fill_verify_state_seed(struct thread_data *);
568 extern void add_job_opts(const char **, int);
569 extern int ioengine_load(struct thread_data *);
570 extern bool parse_dryrun(void);
571 extern int fio_running_or_pending_io_threads(void);
572 extern int fio_set_fd_nonblocking(int, const char *);
573 extern void sig_show_status(int sig);
574 extern struct thread_data *get_global_options(void);
576 extern uintptr_t page_mask;
577 extern uintptr_t page_size;
578 extern int initialize_fio(char *envp[]);
579 extern void deinitialize_fio(void);
581 #define FIO_GETOPT_JOB 0x89000000
582 #define FIO_GETOPT_IOENGINE 0x98000000
583 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
588 extern void print_thread_status(void);
589 extern void print_status_init(int);
590 extern char *fio_uint_to_kmg(unsigned int val);
593 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
594 * will never back again. It may cycle between running/verififying/fsyncing.
595 * Once the thread reaches TD_EXITED, it is just waiting for the core to
615 #define TD_ENG_FLAG_SHIFT 17
616 #define TD_ENG_FLAG_MASK ((1U << 17) - 1)
618 static inline void td_set_ioengine_flags(struct thread_data *td)
620 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
621 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
624 static inline bool td_ioengine_flagged(struct thread_data *td,
625 enum fio_ioengine_flags flags)
627 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
630 extern void td_set_runstate(struct thread_data *, int);
631 extern int td_bump_runstate(struct thread_data *, int);
632 extern void td_restore_runstate(struct thread_data *, int);
633 extern const char *runstate_to_name(int runstate);
636 * Allow 60 seconds for a job to quit on its own, otherwise reap with
639 #define FIO_REAP_TIMEOUT 300
641 #define TERMINATE_ALL (-1U)
642 extern void fio_terminate_threads(unsigned int);
643 extern void fio_mark_td_terminate(struct thread_data *);
648 extern int __must_check fio_pin_memory(struct thread_data *);
649 extern void fio_unpin_memory(struct thread_data *);
650 extern int __must_check allocate_io_mem(struct thread_data *);
651 extern void free_io_mem(struct thread_data *);
652 extern void free_threads_shm(void);
655 #define PTR_ALIGN(ptr, mask) \
656 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
660 * Reset stats after ramp time completes
662 extern void reset_all_stats(struct thread_data *);
664 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
665 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
666 struct timespec *comp_time);
669 * Latency target helpers
671 extern void lat_target_check(struct thread_data *);
672 extern void lat_target_init(struct thread_data *);
673 extern void lat_target_reset(struct thread_data *);
676 * Iterates all threads/processes within all the defined jobs
678 #define for_each_td(td, i) \
679 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
680 #define for_each_file(td, f, i) \
681 if ((td)->files_index) \
682 for ((i) = 0, (f) = (td)->files[0]; \
683 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
686 #define fio_assert(td, cond) do { \
689 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
690 td_set_runstate((td), TD_EXITED); \
691 (td)->error = EFAULT; \
696 static inline bool fio_fill_issue_time(struct thread_data *td)
698 if (td->o.read_iolog_file ||
699 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
705 static inline bool option_check_rate(struct thread_data *td, enum fio_ddir ddir)
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,
722 return (td->flags & TD_F_CHECK_RATE) != 0;
725 static inline bool should_check_rate(struct thread_data *td)
727 if (__should_check_rate(td, DDIR_READ) && td->bytes_done[DDIR_READ])
729 if (__should_check_rate(td, DDIR_WRITE) && td->bytes_done[DDIR_WRITE])
731 if (__should_check_rate(td, DDIR_TRIM) && td->bytes_done[DDIR_TRIM])
737 static inline unsigned int td_max_bs(struct thread_data *td)
741 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
742 return max(td->o.max_bs[DDIR_TRIM], max_bs);
745 static inline unsigned int td_min_bs(struct thread_data *td)
749 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
750 return min(td->o.min_bs[DDIR_TRIM], min_bs);
753 static inline bool td_async_processing(struct thread_data *td)
755 return (td->flags & TD_F_NEED_LOCK) != 0;
759 * We currently only need to do locking if we have verifier threads
760 * accessing our internal structures too
762 static inline void td_io_u_lock(struct thread_data *td)
764 if (td_async_processing(td))
765 pthread_mutex_lock(&td->io_u_lock);
768 static inline void td_io_u_unlock(struct thread_data *td)
770 if (td_async_processing(td))
771 pthread_mutex_unlock(&td->io_u_lock);
774 static inline void td_io_u_free_notify(struct thread_data *td)
776 if (td_async_processing(td))
777 pthread_cond_signal(&td->free_cond);
780 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
783 if (!td_async_processing(td))
786 __sync_fetch_and_and(flags, ~value);
789 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
792 if (!td_async_processing(td))
795 __sync_fetch_and_or(flags, value);
798 extern const char *fio_get_arch_string(int);
799 extern const char *fio_get_os_string(int);
802 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
803 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
807 __FIO_OUTPUT_TERSE = 0,
808 __FIO_OUTPUT_JSON = 1,
809 __FIO_OUTPUT_NORMAL = 2,
810 __FIO_OUTPUT_JSON_PLUS = 3,
813 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
814 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
815 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
816 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
820 FIO_RAND_DIST_RANDOM = 0,
822 FIO_RAND_DIST_PARETO,
825 FIO_RAND_DIST_ZONED_ABS,
828 #define FIO_DEF_ZIPF 1.1
829 #define FIO_DEF_PARETO 0.2
832 FIO_RAND_GEN_TAUSWORTHE = 0,
834 FIO_RAND_GEN_TAUSWORTHE64,
842 extern void exec_trigger(const char *);
843 extern void check_trigger_file(void);