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 = 1U << 0,
76 TD_F_TRIM_BACKLOG = 1U << 1,
77 TD_F_READ_IOLOG = 1U << 2,
78 TD_F_REFILL_BUFFERS = 1U << 3,
79 TD_F_SCRAMBLE_BUFFERS = 1U << 4,
80 TD_F_VER_NONE = 1U << 5,
81 TD_F_PROFILE_OPS = 1U << 6,
82 TD_F_COMPRESS = 1U << 7,
83 TD_F_RESERVED = 1U << 8, /* not used */
84 TD_F_COMPRESS_LOG = 1U << 9,
85 TD_F_VSTATE_SAVED = 1U << 10,
86 TD_F_NEED_LOCK = 1U << 11,
87 TD_F_CHILD = 1U << 12,
88 TD_F_NO_PROGRESS = 1U << 13,
89 TD_F_REGROW_LOGS = 1U << 14,
100 FIO_RAND_FILE_SIZE_OFF,
103 FIO_RAND_SEQ_RAND_READ_OFF,
104 FIO_RAND_SEQ_RAND_WRITE_OFF,
105 FIO_RAND_SEQ_RAND_TRIM_OFF,
106 FIO_RAND_START_DELAY,
108 FIO_RAND_POISSON_OFF,
110 FIO_RAND_POISSON2_OFF,
111 FIO_RAND_POISSON3_OFF,
119 RATE_PROCESS_LINEAR = 0,
120 RATE_PROCESS_POISSON = 1,
131 * Per-thread/process specific data. Only used for the network client
134 void sk_out_assign(struct sk_out *);
135 void sk_out_drop(void);
137 struct zone_split_index {
139 uint8_t size_perc_prev;
143 * This describes a single thread/process executing a fio job.
146 struct flist_head opt_list;
148 struct thread_options o;
151 unsigned int thread_number;
152 unsigned int subjob_number;
153 unsigned int groupid;
154 struct thread_stat ts __attribute__ ((aligned(8)));
158 struct io_log *slat_log;
159 struct io_log *clat_log;
160 struct io_log *clat_hist_log;
161 struct io_log *lat_log;
162 struct io_log *bw_log;
163 struct io_log *iops_log;
165 struct workqueue log_compress_wq;
167 struct thread_data *parent;
169 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
170 struct timespec bw_sample_time;
172 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
173 struct timespec iops_sample_time;
175 volatile int update_rusage;
176 struct fio_mutex *rusage_sem;
177 struct rusage ru_start;
178 struct rusage ru_end;
180 struct fio_file **files;
181 unsigned char *file_locks;
182 unsigned int files_size;
183 unsigned int files_index;
184 unsigned int nr_open_files;
185 unsigned int nr_done_files;
186 unsigned int nr_normal_files;
188 unsigned int next_file;
189 struct frand_state next_file_state;
192 struct zipf_state next_file_zipf;
193 struct gauss_state next_file_gauss;
206 size_t orig_buffer_size;
207 volatile int terminate;
208 volatile int runstate;
209 unsigned int last_was_sync;
210 enum fio_ddir last_ddir;
217 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
219 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
220 struct frand_state verify_state;
221 struct frand_state trim_state;
222 struct frand_state delay_state;
224 struct frand_state buf_state;
225 struct frand_state buf_state_prev;
226 struct frand_state dedupe_state;
227 struct frand_state zone_state;
229 struct zone_split_index **zone_state_index;
231 unsigned int verify_batch;
232 unsigned int trim_batch;
234 struct thread_io_list *vstate;
239 * IO engine hooks, contains everything needed to submit an io_u
240 * to any of the available IO engines.
242 struct ioengine_ops *io_ops;
246 * IO engine private data and dlhandle.
249 void *io_ops_dlhandle;
252 * Queue depth of io_u's that fio MIGHT do
254 unsigned int cur_depth;
257 * io_u's about to be committed
259 unsigned int io_u_queued;
262 * io_u's submitted but not completed yet
264 unsigned int io_u_in_flight;
267 * List of free and busy io_u's
269 struct io_u_ring io_u_requeues;
270 struct io_u_queue io_u_freelist;
271 struct io_u_queue io_u_all;
272 pthread_mutex_t io_u_lock;
273 pthread_cond_t free_cond;
276 * async verify offload
278 struct flist_head verify_list;
279 pthread_t *verify_threads;
280 unsigned int nr_verify_threads;
281 pthread_cond_t verify_cond;
282 int verify_thread_exit;
287 uint64_t rate_bps[DDIR_RWDIR_CNT];
288 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
289 unsigned long rate_bytes[DDIR_RWDIR_CNT];
290 unsigned long rate_blocks[DDIR_RWDIR_CNT];
291 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
292 struct timespec lastrate[DDIR_RWDIR_CNT];
293 int64_t last_usec[DDIR_RWDIR_CNT];
294 struct frand_state poisson_state[DDIR_RWDIR_CNT];
297 * Enforced rate submission/completion workqueue
299 struct workqueue io_wq;
301 uint64_t total_io_size;
302 uint64_t fill_device_size;
307 uint64_t io_issues[DDIR_RWDIR_CNT];
308 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
314 uint64_t io_blocks[DDIR_RWDIR_CNT];
315 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
316 uint64_t io_bytes[DDIR_RWDIR_CNT];
317 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
318 uint64_t io_skip_bytes;
320 struct fio_mutex *mutex;
321 uint64_t bytes_done[DDIR_RWDIR_CNT];
324 * State for random io, a bitmap of blocks done vs not done
326 struct frand_state random_state;
328 struct timespec start; /* start of this loop */
329 struct timespec epoch; /* time job was started */
330 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
331 struct timespec last_issue;
333 struct timespec ts_cache;
334 struct timespec terminate_time;
335 unsigned int ts_cache_nr;
336 unsigned int ts_cache_mask;
337 unsigned int ramp_time_over;
340 * Time since last latency_window was started
342 struct timespec latency_ts;
343 unsigned int latency_qd;
344 unsigned int latency_qd_high;
345 unsigned int latency_qd_low;
346 unsigned int latency_failed;
347 uint64_t latency_ios;
351 * read/write mixed workload state
353 struct frand_state rwmix_state;
354 unsigned long rwmix_issues;
355 enum fio_ddir rwmix_ddir;
356 unsigned int ddir_seq_nr;
359 * rand/seq mixed workload state
361 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
364 * IO history logs for verification. We use a tree for sorting,
365 * if we are overwriting. Otherwise just use a fifo.
367 struct rb_root io_hist_tree;
368 struct flist_head io_hist_list;
369 unsigned long io_hist_len;
374 struct flist_head io_log_list;
377 * For tracking/handling discards
379 struct flist_head trim_list;
380 unsigned long trim_entries;
382 struct flist_head next_rand_list;
385 * for fileservice, how often to switch to a new file
387 unsigned int file_service_nr;
388 unsigned int file_service_left;
389 struct fio_file *file_service_file;
391 unsigned int sync_file_range_nr;
394 * For generating file sizes
396 struct frand_state file_size_state;
401 unsigned int total_err_count;
404 struct fio_flow *flow;
407 * Can be overloaded by profiles
409 struct prof_io_ops prof_io_ops;
414 struct steadystate_data ss;
416 char verror[FIO_VERROR_SIZE];
420 * for GPU memory management
426 CUdeviceptr dev_mem_ptr;
432 * when should interactive ETA output be generated
440 #define __td_verror(td, err, msg, func) \
442 unsigned int ____e = (err); \
445 (td)->error = ____e; \
446 if (!(td)->first_error) \
447 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
451 #define td_clear_error(td) do { \
454 (td)->parent->error = 0; \
457 #define td_verror(td, err, func) do { \
458 __td_verror((td), (err), strerror((err)), (func)); \
460 __td_verror((td)->parent, (err), strerror((err)), (func)); \
463 #define td_vmsg(td, err, msg, func) do { \
464 __td_verror((td), (err), (msg), (func)); \
466 __td_verror((td)->parent, (err), (msg), (func)); \
469 #define __fio_stringify_1(x) #x
470 #define __fio_stringify(x) __fio_stringify_1(x)
472 extern int exitall_on_terminate;
473 extern unsigned int thread_number;
474 extern unsigned int stat_number;
477 extern int output_format;
478 extern int append_terse_output;
479 extern int temp_stall_ts;
480 extern uintptr_t page_mask, page_size;
481 extern int read_only;
482 extern int eta_print;
483 extern int eta_new_line;
484 extern unsigned long done_secs;
485 extern int fio_gtod_offload;
486 extern int fio_gtod_cpu;
487 extern enum fio_cs fio_clock_source;
488 extern int fio_clock_source_set;
489 extern int warnings_fatal;
490 extern int terse_version;
491 extern int is_backend;
492 extern int nr_clients;
493 extern int log_syslog;
494 extern int status_interval;
495 extern const char fio_version_string[];
496 extern char *trigger_file;
497 extern char *trigger_cmd;
498 extern char *trigger_remote_cmd;
499 extern long long trigger_timeout;
500 extern char *aux_path;
502 extern struct thread_data *threads;
504 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
506 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
509 #define REAL_MAX_JOBS 4096
511 static inline int should_fsync(struct thread_data *td)
513 if (td->last_was_sync)
515 if (td_write(td) || td->o.override_sync)
522 * Init/option functions
524 extern int __must_check fio_init_options(void);
525 extern int __must_check parse_options(int, char **);
526 extern int parse_jobs_ini(char *, int, int, int);
527 extern int parse_cmd_line(int, char **, int);
528 extern int fio_backend(struct sk_out *);
529 extern void reset_fio_state(void);
530 extern void clear_io_state(struct thread_data *, int);
531 extern int fio_options_parse(struct thread_data *, char **, int);
532 extern void fio_keywords_init(void);
533 extern void fio_keywords_exit(void);
534 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
535 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
536 extern void fio_fill_default_options(struct thread_data *);
537 extern int fio_show_option_help(const char *);
538 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
539 extern void fio_options_dup_and_init(struct option *);
540 extern void fio_options_mem_dupe(struct thread_data *);
541 extern void td_fill_rand_seeds(struct thread_data *);
542 extern void td_fill_verify_state_seed(struct thread_data *);
543 extern void add_job_opts(const char **, int);
544 extern int ioengine_load(struct thread_data *);
545 extern bool parse_dryrun(void);
546 extern int fio_running_or_pending_io_threads(void);
547 extern int fio_set_fd_nonblocking(int, const char *);
548 extern void sig_show_status(int sig);
549 extern struct thread_data *get_global_options(void);
551 extern uintptr_t page_mask;
552 extern uintptr_t page_size;
553 extern int initialize_fio(char *envp[]);
554 extern void deinitialize_fio(void);
556 #define FIO_GETOPT_JOB 0x89000000
557 #define FIO_GETOPT_IOENGINE 0x98000000
558 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
563 extern void print_thread_status(void);
564 extern void print_status_init(int);
565 extern char *fio_uint_to_kmg(unsigned int val);
568 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
569 * will never back again. It may cycle between running/verififying/fsyncing.
570 * Once the thread reaches TD_EXITED, it is just waiting for the core to
590 #define TD_ENG_FLAG_SHIFT 16
591 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
593 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
595 return (enum fio_ioengine_flags)
596 ((td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK);
599 static inline void td_set_ioengine_flags(struct thread_data *td)
601 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
602 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
605 static inline bool td_ioengine_flagged(struct thread_data *td,
606 enum fio_ioengine_flags flags)
608 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
611 extern void td_set_runstate(struct thread_data *, int);
612 extern int td_bump_runstate(struct thread_data *, int);
613 extern void td_restore_runstate(struct thread_data *, int);
614 extern const char *runstate_to_name(int runstate);
617 * Allow 60 seconds for a job to quit on its own, otherwise reap with
620 #define FIO_REAP_TIMEOUT 300
622 #define TERMINATE_ALL (-1U)
623 extern void fio_terminate_threads(unsigned int);
624 extern void fio_mark_td_terminate(struct thread_data *);
629 extern int __must_check fio_pin_memory(struct thread_data *);
630 extern void fio_unpin_memory(struct thread_data *);
631 extern int __must_check allocate_io_mem(struct thread_data *);
632 extern void free_io_mem(struct thread_data *);
633 extern void free_threads_shm(void);
636 #define PTR_ALIGN(ptr, mask) \
637 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
641 * Reset stats after ramp time completes
643 extern void reset_all_stats(struct thread_data *);
645 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
646 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
647 struct timespec *comp_time);
650 * Latency target helpers
652 extern void lat_target_check(struct thread_data *);
653 extern void lat_target_init(struct thread_data *);
654 extern void lat_target_reset(struct thread_data *);
657 * Iterates all threads/processes within all the defined jobs
659 #define for_each_td(td, i) \
660 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
661 #define for_each_file(td, f, i) \
662 if ((td)->files_index) \
663 for ((i) = 0, (f) = (td)->files[0]; \
664 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
667 #define fio_assert(td, cond) do { \
670 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
671 td_set_runstate((td), TD_EXITED); \
672 (td)->error = EFAULT; \
677 static inline bool fio_fill_issue_time(struct thread_data *td)
679 if (td->o.read_iolog_file ||
680 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
686 static inline bool __should_check_rate(struct thread_data *td,
689 struct thread_options *o = &td->o;
692 * If some rate setting was given, we need to check it
694 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
695 o->rate_iops_min[ddir])
701 static inline bool should_check_rate(struct thread_data *td)
703 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
705 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
707 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
713 static inline unsigned int td_max_bs(struct thread_data *td)
717 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
718 return max(td->o.max_bs[DDIR_TRIM], max_bs);
721 static inline unsigned int td_min_bs(struct thread_data *td)
725 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
726 return min(td->o.min_bs[DDIR_TRIM], min_bs);
729 static inline bool td_async_processing(struct thread_data *td)
731 return (td->flags & TD_F_NEED_LOCK) != 0;
735 * We currently only need to do locking if we have verifier threads
736 * accessing our internal structures too
738 static inline void td_io_u_lock(struct thread_data *td)
740 if (td_async_processing(td))
741 pthread_mutex_lock(&td->io_u_lock);
744 static inline void td_io_u_unlock(struct thread_data *td)
746 if (td_async_processing(td))
747 pthread_mutex_unlock(&td->io_u_lock);
750 static inline void td_io_u_free_notify(struct thread_data *td)
752 if (td_async_processing(td))
753 pthread_cond_signal(&td->free_cond);
756 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
759 if (!td_async_processing(td))
762 __sync_fetch_and_and(flags, ~value);
765 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
768 if (!td_async_processing(td))
771 __sync_fetch_and_or(flags, value);
774 extern const char *fio_get_arch_string(int);
775 extern const char *fio_get_os_string(int);
778 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
779 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
783 __FIO_OUTPUT_TERSE = 0,
784 __FIO_OUTPUT_JSON = 1,
785 __FIO_OUTPUT_NORMAL = 2,
786 __FIO_OUTPUT_JSON_PLUS = 3,
789 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
790 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
791 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
792 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
796 FIO_RAND_DIST_RANDOM = 0,
798 FIO_RAND_DIST_PARETO,
803 #define FIO_DEF_ZIPF 1.1
804 #define FIO_DEF_PARETO 0.2
807 FIO_RAND_GEN_TAUSWORTHE = 0,
809 FIO_RAND_GEN_TAUSWORTHE64,
817 extern void exec_trigger(const char *);
818 extern void check_trigger_file(void);