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,
90 __TD_F_LAST, /* not a real bit, keep last */
94 TD_F_VER_BACKLOG = 1U << __TD_F_VER_BACKLOG,
95 TD_F_TRIM_BACKLOG = 1U << __TD_F_TRIM_BACKLOG,
96 TD_F_READ_IOLOG = 1U << __TD_F_READ_IOLOG,
97 TD_F_REFILL_BUFFERS = 1U << __TD_F_REFILL_BUFFERS,
98 TD_F_SCRAMBLE_BUFFERS = 1U << __TD_F_SCRAMBLE_BUFFERS,
99 TD_F_VER_NONE = 1U << __TD_F_VER_NONE,
100 TD_F_PROFILE_OPS = 1U << __TD_F_PROFILE_OPS,
101 TD_F_COMPRESS = 1U << __TD_F_COMPRESS,
102 TD_F_COMPRESS_LOG = 1U << __TD_F_COMPRESS_LOG,
103 TD_F_VSTATE_SAVED = 1U << __TD_F_VSTATE_SAVED,
104 TD_F_NEED_LOCK = 1U << __TD_F_NEED_LOCK,
105 TD_F_CHILD = 1U << __TD_F_CHILD,
106 TD_F_NO_PROGRESS = 1U << __TD_F_NO_PROGRESS,
107 TD_F_REGROW_LOGS = 1U << __TD_F_REGROW_LOGS,
108 TD_F_MMAP_KEEP = 1U << __TD_F_MMAP_KEEP,
119 FIO_RAND_FILE_SIZE_OFF,
122 FIO_RAND_SEQ_RAND_READ_OFF,
123 FIO_RAND_SEQ_RAND_WRITE_OFF,
124 FIO_RAND_SEQ_RAND_TRIM_OFF,
125 FIO_RAND_START_DELAY,
127 FIO_RAND_POISSON_OFF,
129 FIO_RAND_POISSON2_OFF,
130 FIO_RAND_POISSON3_OFF,
138 RATE_PROCESS_LINEAR = 0,
139 RATE_PROCESS_POISSON = 1,
150 * Per-thread/process specific data. Only used for the network client
153 void sk_out_assign(struct sk_out *);
154 void sk_out_drop(void);
156 struct zone_split_index {
158 uint8_t size_perc_prev;
162 * This describes a single thread/process executing a fio job.
165 struct flist_head opt_list;
167 struct thread_options o;
170 unsigned int thread_number;
171 unsigned int subjob_number;
172 unsigned int groupid;
173 struct thread_stat ts __attribute__ ((aligned(8)));
177 struct io_log *slat_log;
178 struct io_log *clat_log;
179 struct io_log *clat_hist_log;
180 struct io_log *lat_log;
181 struct io_log *bw_log;
182 struct io_log *iops_log;
184 struct workqueue log_compress_wq;
186 struct thread_data *parent;
188 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
189 struct timespec bw_sample_time;
191 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
192 struct timespec iops_sample_time;
194 volatile int update_rusage;
195 struct fio_mutex *rusage_sem;
196 struct rusage ru_start;
197 struct rusage ru_end;
199 struct fio_file **files;
200 unsigned char *file_locks;
201 unsigned int files_size;
202 unsigned int files_index;
203 unsigned int nr_open_files;
204 unsigned int nr_done_files;
205 unsigned int nr_normal_files;
207 unsigned int next_file;
208 struct frand_state next_file_state;
211 struct zipf_state next_file_zipf;
212 struct gauss_state next_file_gauss;
225 size_t orig_buffer_size;
226 volatile int terminate;
227 volatile int runstate;
228 unsigned int last_was_sync;
229 enum fio_ddir last_ddir;
236 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
238 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
239 struct frand_state verify_state;
240 struct frand_state trim_state;
241 struct frand_state delay_state;
243 struct frand_state buf_state;
244 struct frand_state buf_state_prev;
245 struct frand_state dedupe_state;
246 struct frand_state zone_state;
248 struct zone_split_index **zone_state_index;
250 unsigned int verify_batch;
251 unsigned int trim_batch;
253 struct thread_io_list *vstate;
258 * IO engine hooks, contains everything needed to submit an io_u
259 * to any of the available IO engines.
261 struct ioengine_ops *io_ops;
265 * IO engine private data and dlhandle.
268 void *io_ops_dlhandle;
271 * Queue depth of io_u's that fio MIGHT do
273 unsigned int cur_depth;
276 * io_u's about to be committed
278 unsigned int io_u_queued;
281 * io_u's submitted but not completed yet
283 unsigned int io_u_in_flight;
286 * List of free and busy io_u's
288 struct io_u_ring io_u_requeues;
289 struct io_u_queue io_u_freelist;
290 struct io_u_queue io_u_all;
291 pthread_mutex_t io_u_lock;
292 pthread_cond_t free_cond;
295 * async verify offload
297 struct flist_head verify_list;
298 pthread_t *verify_threads;
299 unsigned int nr_verify_threads;
300 pthread_cond_t verify_cond;
301 int verify_thread_exit;
306 uint64_t rate_bps[DDIR_RWDIR_CNT];
307 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
308 unsigned long rate_bytes[DDIR_RWDIR_CNT];
309 unsigned long rate_blocks[DDIR_RWDIR_CNT];
310 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
311 struct timespec lastrate[DDIR_RWDIR_CNT];
312 int64_t last_usec[DDIR_RWDIR_CNT];
313 struct frand_state poisson_state[DDIR_RWDIR_CNT];
316 * Enforced rate submission/completion workqueue
318 struct workqueue io_wq;
320 uint64_t total_io_size;
321 uint64_t fill_device_size;
326 uint64_t io_issues[DDIR_RWDIR_CNT];
327 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
333 uint64_t io_blocks[DDIR_RWDIR_CNT];
334 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
335 uint64_t io_bytes[DDIR_RWDIR_CNT];
336 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
337 uint64_t io_skip_bytes;
339 struct fio_mutex *mutex;
340 uint64_t bytes_done[DDIR_RWDIR_CNT];
343 * State for random io, a bitmap of blocks done vs not done
345 struct frand_state random_state;
347 struct timespec start; /* start of this loop */
348 struct timespec epoch; /* time job was started */
349 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
350 struct timespec last_issue;
352 struct timespec ts_cache;
353 struct timespec terminate_time;
354 unsigned int ts_cache_nr;
355 unsigned int ts_cache_mask;
359 * Time since last latency_window was started
361 struct timespec latency_ts;
362 unsigned int latency_qd;
363 unsigned int latency_qd_high;
364 unsigned int latency_qd_low;
365 unsigned int latency_failed;
366 uint64_t latency_ios;
370 * read/write mixed workload state
372 struct frand_state rwmix_state;
373 unsigned long rwmix_issues;
374 enum fio_ddir rwmix_ddir;
375 unsigned int ddir_seq_nr;
378 * rand/seq mixed workload state
380 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
383 * IO history logs for verification. We use a tree for sorting,
384 * if we are overwriting. Otherwise just use a fifo.
386 struct rb_root io_hist_tree;
387 struct flist_head io_hist_list;
388 unsigned long io_hist_len;
393 struct flist_head io_log_list;
396 * For tracking/handling discards
398 struct flist_head trim_list;
399 unsigned long trim_entries;
401 struct flist_head next_rand_list;
404 * for fileservice, how often to switch to a new file
406 unsigned int file_service_nr;
407 unsigned int file_service_left;
408 struct fio_file *file_service_file;
410 unsigned int sync_file_range_nr;
413 * For generating file sizes
415 struct frand_state file_size_state;
420 unsigned int total_err_count;
423 struct fio_flow *flow;
426 * Can be overloaded by profiles
428 struct prof_io_ops prof_io_ops;
433 struct steadystate_data ss;
435 char verror[FIO_VERROR_SIZE];
439 * for GPU memory management
445 CUdeviceptr dev_mem_ptr;
451 * when should interactive ETA output be generated
459 #define __td_verror(td, err, msg, func) \
461 unsigned int ____e = (err); \
464 (td)->error = ____e; \
465 if (!(td)->first_error) \
466 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
470 #define td_clear_error(td) do { \
473 (td)->parent->error = 0; \
476 #define td_verror(td, err, func) do { \
477 __td_verror((td), (err), strerror((err)), (func)); \
479 __td_verror((td)->parent, (err), strerror((err)), (func)); \
482 #define td_vmsg(td, err, msg, func) do { \
483 __td_verror((td), (err), (msg), (func)); \
485 __td_verror((td)->parent, (err), (msg), (func)); \
488 #define __fio_stringify_1(x) #x
489 #define __fio_stringify(x) __fio_stringify_1(x)
491 extern int exitall_on_terminate;
492 extern unsigned int thread_number;
493 extern unsigned int stat_number;
496 extern int output_format;
497 extern int append_terse_output;
498 extern int temp_stall_ts;
499 extern uintptr_t page_mask, page_size;
500 extern int read_only;
501 extern int eta_print;
502 extern int eta_new_line;
503 extern unsigned long done_secs;
504 extern int fio_gtod_offload;
505 extern int fio_gtod_cpu;
506 extern enum fio_cs fio_clock_source;
507 extern int fio_clock_source_set;
508 extern int warnings_fatal;
509 extern int terse_version;
510 extern int is_backend;
511 extern int nr_clients;
512 extern int log_syslog;
513 extern int status_interval;
514 extern const char fio_version_string[];
515 extern char *trigger_file;
516 extern char *trigger_cmd;
517 extern char *trigger_remote_cmd;
518 extern long long trigger_timeout;
519 extern char *aux_path;
521 extern struct thread_data *threads;
523 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
525 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
528 #define REAL_MAX_JOBS 4096
530 static inline int should_fsync(struct thread_data *td)
532 if (td->last_was_sync)
534 if (td_write(td) || td->o.override_sync)
541 * Init/option functions
543 extern int __must_check fio_init_options(void);
544 extern int __must_check parse_options(int, char **);
545 extern int parse_jobs_ini(char *, int, int, int);
546 extern int parse_cmd_line(int, char **, int);
547 extern int fio_backend(struct sk_out *);
548 extern void reset_fio_state(void);
549 extern void clear_io_state(struct thread_data *, int);
550 extern int fio_options_parse(struct thread_data *, char **, int);
551 extern void fio_keywords_init(void);
552 extern void fio_keywords_exit(void);
553 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
554 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
555 extern void fio_fill_default_options(struct thread_data *);
556 extern int fio_show_option_help(const char *);
557 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
558 extern void fio_options_dup_and_init(struct option *);
559 extern void fio_options_mem_dupe(struct thread_data *);
560 extern void td_fill_rand_seeds(struct thread_data *);
561 extern void td_fill_verify_state_seed(struct thread_data *);
562 extern void add_job_opts(const char **, int);
563 extern int ioengine_load(struct thread_data *);
564 extern bool parse_dryrun(void);
565 extern int fio_running_or_pending_io_threads(void);
566 extern int fio_set_fd_nonblocking(int, const char *);
567 extern void sig_show_status(int sig);
568 extern struct thread_data *get_global_options(void);
570 extern uintptr_t page_mask;
571 extern uintptr_t page_size;
572 extern int initialize_fio(char *envp[]);
573 extern void deinitialize_fio(void);
575 #define FIO_GETOPT_JOB 0x89000000
576 #define FIO_GETOPT_IOENGINE 0x98000000
577 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
582 extern void print_thread_status(void);
583 extern void print_status_init(int);
584 extern char *fio_uint_to_kmg(unsigned int val);
587 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
588 * will never back again. It may cycle between running/verififying/fsyncing.
589 * Once the thread reaches TD_EXITED, it is just waiting for the core to
609 #define TD_ENG_FLAG_SHIFT 16
610 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
612 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
614 return (enum fio_ioengine_flags)
615 ((td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK);
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 __should_check_rate(struct thread_data *td,
708 struct thread_options *o = &td->o;
711 * If some rate setting was given, we need to check it
713 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
714 o->rate_iops_min[ddir])
720 static inline bool should_check_rate(struct thread_data *td)
722 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
724 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
726 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
732 static inline unsigned int td_max_bs(struct thread_data *td)
736 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
737 return max(td->o.max_bs[DDIR_TRIM], max_bs);
740 static inline unsigned int td_min_bs(struct thread_data *td)
744 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
745 return min(td->o.min_bs[DDIR_TRIM], min_bs);
748 static inline bool td_async_processing(struct thread_data *td)
750 return (td->flags & TD_F_NEED_LOCK) != 0;
754 * We currently only need to do locking if we have verifier threads
755 * accessing our internal structures too
757 static inline void td_io_u_lock(struct thread_data *td)
759 if (td_async_processing(td))
760 pthread_mutex_lock(&td->io_u_lock);
763 static inline void td_io_u_unlock(struct thread_data *td)
765 if (td_async_processing(td))
766 pthread_mutex_unlock(&td->io_u_lock);
769 static inline void td_io_u_free_notify(struct thread_data *td)
771 if (td_async_processing(td))
772 pthread_cond_signal(&td->free_cond);
775 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
778 if (!td_async_processing(td))
781 __sync_fetch_and_and(flags, ~value);
784 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
787 if (!td_async_processing(td))
790 __sync_fetch_and_or(flags, value);
793 extern const char *fio_get_arch_string(int);
794 extern const char *fio_get_os_string(int);
797 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
798 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
802 __FIO_OUTPUT_TERSE = 0,
803 __FIO_OUTPUT_JSON = 1,
804 __FIO_OUTPUT_NORMAL = 2,
805 __FIO_OUTPUT_JSON_PLUS = 3,
808 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
809 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
810 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
811 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
815 FIO_RAND_DIST_RANDOM = 0,
817 FIO_RAND_DIST_PARETO,
822 #define FIO_DEF_ZIPF 1.1
823 #define FIO_DEF_PARETO 0.2
826 FIO_RAND_GEN_TAUSWORTHE = 0,
828 FIO_RAND_GEN_TAUSWORTHE64,
836 extern void exec_trigger(const char *);
837 extern void check_trigger_file(void);