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,
98 FIO_RAND_FILE_SIZE_OFF,
101 FIO_RAND_SEQ_RAND_READ_OFF,
102 FIO_RAND_SEQ_RAND_WRITE_OFF,
103 FIO_RAND_SEQ_RAND_TRIM_OFF,
104 FIO_RAND_START_DELAY,
106 FIO_RAND_POISSON_OFF,
108 FIO_RAND_POISSON2_OFF,
109 FIO_RAND_POISSON3_OFF,
117 RATE_PROCESS_LINEAR = 0,
118 RATE_PROCESS_POISSON = 1,
129 * Per-thread/process specific data. Only used for the network client
132 void sk_out_assign(struct sk_out *);
133 void sk_out_drop(void);
135 struct zone_split_index {
137 uint8_t size_perc_prev;
141 * This describes a single thread/process executing a fio job.
144 struct flist_head opt_list;
146 struct thread_options o;
149 unsigned int thread_number;
150 unsigned int subjob_number;
151 unsigned int groupid;
152 struct thread_stat ts;
156 struct io_log *slat_log;
157 struct io_log *clat_log;
158 struct io_log *clat_hist_log;
159 struct io_log *lat_log;
160 struct io_log *bw_log;
161 struct io_log *iops_log;
163 struct workqueue log_compress_wq;
165 struct thread_data *parent;
167 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
168 struct timeval bw_sample_time;
170 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
171 struct timeval iops_sample_time;
173 volatile int update_rusage;
174 struct fio_mutex *rusage_sem;
175 struct rusage ru_start;
176 struct rusage ru_end;
178 struct fio_file **files;
179 unsigned char *file_locks;
180 unsigned int files_size;
181 unsigned int files_index;
182 unsigned int nr_open_files;
183 unsigned int nr_done_files;
184 unsigned int nr_normal_files;
186 unsigned int next_file;
187 struct frand_state next_file_state;
190 struct zipf_state next_file_zipf;
191 struct gauss_state next_file_gauss;
204 size_t orig_buffer_size;
205 volatile int terminate;
206 volatile int runstate;
207 unsigned int last_was_sync;
208 enum fio_ddir last_ddir;
215 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
217 struct frand_state bsrange_state;
218 struct frand_state verify_state;
219 struct frand_state trim_state;
220 struct frand_state delay_state;
222 struct frand_state buf_state;
223 struct frand_state buf_state_prev;
224 struct frand_state dedupe_state;
225 struct frand_state zone_state;
227 struct zone_split_index **zone_state_index;
229 unsigned int verify_batch;
230 unsigned int trim_batch;
232 struct thread_io_list *vstate;
237 * IO engine hooks, contains everything needed to submit an io_u
238 * to any of the available IO engines.
240 struct ioengine_ops *io_ops;
244 * IO engine private data and dlhandle.
247 void *io_ops_dlhandle;
250 * Queue depth of io_u's that fio MIGHT do
252 unsigned int cur_depth;
255 * io_u's about to be committed
257 unsigned int io_u_queued;
260 * io_u's submitted but not completed yet
262 unsigned int io_u_in_flight;
265 * List of free and busy io_u's
267 struct io_u_ring io_u_requeues;
268 struct io_u_queue io_u_freelist;
269 struct io_u_queue io_u_all;
270 pthread_mutex_t io_u_lock;
271 pthread_cond_t free_cond;
274 * async verify offload
276 struct flist_head verify_list;
277 pthread_t *verify_threads;
278 unsigned int nr_verify_threads;
279 pthread_cond_t verify_cond;
280 int verify_thread_exit;
285 uint64_t rate_bps[DDIR_RWDIR_CNT];
286 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
287 unsigned long rate_bytes[DDIR_RWDIR_CNT];
288 unsigned long rate_blocks[DDIR_RWDIR_CNT];
289 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
290 struct timeval lastrate[DDIR_RWDIR_CNT];
291 int64_t last_usec[DDIR_RWDIR_CNT];
292 struct frand_state poisson_state[DDIR_RWDIR_CNT];
295 * Enforced rate submission/completion workqueue
297 struct workqueue io_wq;
299 uint64_t total_io_size;
300 uint64_t fill_device_size;
305 uint64_t io_issues[DDIR_RWDIR_CNT];
306 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
312 uint64_t io_blocks[DDIR_RWDIR_CNT];
313 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
314 uint64_t io_bytes[DDIR_RWDIR_CNT];
315 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
316 uint64_t io_skip_bytes;
318 struct fio_mutex *mutex;
319 uint64_t bytes_done[DDIR_RWDIR_CNT];
322 * State for random io, a bitmap of blocks done vs not done
324 struct frand_state random_state;
326 struct timeval start; /* start of this loop */
327 struct timeval epoch; /* time job was started */
328 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
329 struct timeval last_issue;
331 struct timeval tv_cache;
332 struct timeval terminate_time;
333 unsigned int tv_cache_nr;
334 unsigned int tv_cache_mask;
335 unsigned int ramp_time_over;
338 * Time since last latency_window was started
340 struct timeval latency_ts;
341 unsigned int latency_qd;
342 unsigned int latency_qd_high;
343 unsigned int latency_qd_low;
344 unsigned int latency_failed;
345 uint64_t latency_ios;
349 * read/write mixed workload state
351 struct frand_state rwmix_state;
352 unsigned long rwmix_issues;
353 enum fio_ddir rwmix_ddir;
354 unsigned int ddir_seq_nr;
357 * rand/seq mixed workload state
359 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
362 * IO history logs for verification. We use a tree for sorting,
363 * if we are overwriting. Otherwise just use a fifo.
365 struct rb_root io_hist_tree;
366 struct flist_head io_hist_list;
367 unsigned long io_hist_len;
372 struct flist_head io_log_list;
375 * For tracking/handling discards
377 struct flist_head trim_list;
378 unsigned long trim_entries;
380 struct flist_head next_rand_list;
383 * for fileservice, how often to switch to a new file
385 unsigned int file_service_nr;
386 unsigned int file_service_left;
387 struct fio_file *file_service_file;
389 unsigned int sync_file_range_nr;
392 * For generating file sizes
394 struct frand_state file_size_state;
399 unsigned int total_err_count;
402 struct fio_flow *flow;
405 * Can be overloaded by profiles
407 struct prof_io_ops prof_io_ops;
412 struct steadystate_data ss;
414 char verror[FIO_VERROR_SIZE];
418 * for GPU memory management
424 CUdeviceptr dev_mem_ptr;
430 * when should interactive ETA output be generated
438 #define __td_verror(td, err, msg, func) \
440 unsigned int ____e = (err); \
443 (td)->error = ____e; \
444 if (!(td)->first_error) \
445 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
449 #define td_clear_error(td) do { \
452 (td)->parent->error = 0; \
455 #define td_verror(td, err, func) do { \
456 __td_verror((td), (err), strerror((err)), (func)); \
458 __td_verror((td)->parent, (err), strerror((err)), (func)); \
461 #define td_vmsg(td, err, msg, func) do { \
462 __td_verror((td), (err), (msg), (func)); \
464 __td_verror((td)->parent, (err), (msg), (func)); \
467 #define __fio_stringify_1(x) #x
468 #define __fio_stringify(x) __fio_stringify_1(x)
470 extern int exitall_on_terminate;
471 extern unsigned int thread_number;
472 extern unsigned int stat_number;
475 extern int output_format;
476 extern int append_terse_output;
477 extern int temp_stall_ts;
478 extern uintptr_t page_mask, page_size;
479 extern int read_only;
480 extern int eta_print;
481 extern int eta_new_line;
482 extern unsigned long done_secs;
483 extern int fio_gtod_offload;
484 extern int fio_gtod_cpu;
485 extern enum fio_cs fio_clock_source;
486 extern int fio_clock_source_set;
487 extern int warnings_fatal;
488 extern int terse_version;
489 extern int is_backend;
490 extern int nr_clients;
491 extern int log_syslog;
492 extern int status_interval;
493 extern const char fio_version_string[];
494 extern char *trigger_file;
495 extern char *trigger_cmd;
496 extern char *trigger_remote_cmd;
497 extern long long trigger_timeout;
498 extern char *aux_path;
500 extern struct thread_data *threads;
502 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
504 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
507 #define REAL_MAX_JOBS 4096
509 static inline int should_fsync(struct thread_data *td)
511 if (td->last_was_sync)
513 if (td_write(td) || td->o.override_sync)
520 * Init/option functions
522 extern int __must_check fio_init_options(void);
523 extern int __must_check parse_options(int, char **);
524 extern int parse_jobs_ini(char *, int, int, int);
525 extern int parse_cmd_line(int, char **, int);
526 extern int fio_backend(struct sk_out *);
527 extern void reset_fio_state(void);
528 extern void clear_io_state(struct thread_data *, int);
529 extern int fio_options_parse(struct thread_data *, char **, int);
530 extern void fio_keywords_init(void);
531 extern void fio_keywords_exit(void);
532 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
533 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
534 extern void fio_fill_default_options(struct thread_data *);
535 extern int fio_show_option_help(const char *);
536 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
537 extern void fio_options_dup_and_init(struct option *);
538 extern void fio_options_mem_dupe(struct thread_data *);
539 extern void td_fill_rand_seeds(struct thread_data *);
540 extern void td_fill_verify_state_seed(struct thread_data *);
541 extern void add_job_opts(const char **, int);
542 extern int ioengine_load(struct thread_data *);
543 extern bool parse_dryrun(void);
544 extern int fio_running_or_pending_io_threads(void);
545 extern int fio_set_fd_nonblocking(int, const char *);
546 extern void sig_show_status(int sig);
547 extern struct thread_data *get_global_options(void);
549 extern uintptr_t page_mask;
550 extern uintptr_t page_size;
551 extern int initialize_fio(char *envp[]);
552 extern void deinitialize_fio(void);
554 #define FIO_GETOPT_JOB 0x89000000
555 #define FIO_GETOPT_IOENGINE 0x98000000
556 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
561 extern void print_thread_status(void);
562 extern void print_status_init(int);
563 extern char *fio_uint_to_kmg(unsigned int val);
566 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
567 * will never back again. It may cycle between running/verififying/fsyncing.
568 * Once the thread reaches TD_EXITED, it is just waiting for the core to
588 #define TD_ENG_FLAG_SHIFT 16
589 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
591 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
593 return (enum fio_ioengine_flags)
594 ((td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK);
597 static inline void td_set_ioengine_flags(struct thread_data *td)
599 td->flags |= (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
602 static inline bool td_ioengine_flagged(struct thread_data *td,
603 enum fio_ioengine_flags flags)
605 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
608 extern void td_set_runstate(struct thread_data *, int);
609 extern int td_bump_runstate(struct thread_data *, int);
610 extern void td_restore_runstate(struct thread_data *, int);
611 extern const char *runstate_to_name(int runstate);
614 * Allow 60 seconds for a job to quit on its own, otherwise reap with
617 #define FIO_REAP_TIMEOUT 300
619 #define TERMINATE_ALL (-1U)
620 extern void fio_terminate_threads(unsigned int);
621 extern void fio_mark_td_terminate(struct thread_data *);
626 extern int __must_check fio_pin_memory(struct thread_data *);
627 extern void fio_unpin_memory(struct thread_data *);
628 extern int __must_check allocate_io_mem(struct thread_data *);
629 extern void free_io_mem(struct thread_data *);
630 extern void free_threads_shm(void);
633 #define PTR_ALIGN(ptr, mask) \
634 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
638 * Reset stats after ramp time completes
640 extern void reset_all_stats(struct thread_data *);
645 #ifdef FIO_HAVE_BLKTRACE
646 extern int is_blktrace(const char *, int *);
647 extern int load_blktrace(struct thread_data *, const char *, int);
650 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
651 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
652 struct timeval *comp_time);
655 * Latency target helpers
657 extern void lat_target_check(struct thread_data *);
658 extern void lat_target_init(struct thread_data *);
659 extern void lat_target_reset(struct thread_data *);
662 * Iterates all threads/processes within all the defined jobs
664 #define for_each_td(td, i) \
665 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
666 #define for_each_file(td, f, i) \
667 if ((td)->files_index) \
668 for ((i) = 0, (f) = (td)->files[0]; \
669 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
672 #define fio_assert(td, cond) do { \
675 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
676 td_set_runstate((td), TD_EXITED); \
677 (td)->error = EFAULT; \
682 static inline bool fio_fill_issue_time(struct thread_data *td)
684 if (td->o.read_iolog_file ||
685 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
691 static inline bool __should_check_rate(struct thread_data *td,
694 struct thread_options *o = &td->o;
697 * If some rate setting was given, we need to check it
699 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
700 o->rate_iops_min[ddir])
706 static inline bool should_check_rate(struct thread_data *td)
708 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
710 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
712 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
718 static inline unsigned int td_max_bs(struct thread_data *td)
722 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
723 return max(td->o.max_bs[DDIR_TRIM], max_bs);
726 static inline unsigned int td_min_bs(struct thread_data *td)
730 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
731 return min(td->o.min_bs[DDIR_TRIM], min_bs);
734 static inline bool td_async_processing(struct thread_data *td)
736 return (td->flags & TD_F_NEED_LOCK) != 0;
740 * We currently only need to do locking if we have verifier threads
741 * accessing our internal structures too
743 static inline void td_io_u_lock(struct thread_data *td)
745 if (td_async_processing(td))
746 pthread_mutex_lock(&td->io_u_lock);
749 static inline void td_io_u_unlock(struct thread_data *td)
751 if (td_async_processing(td))
752 pthread_mutex_unlock(&td->io_u_lock);
755 static inline void td_io_u_free_notify(struct thread_data *td)
757 if (td_async_processing(td))
758 pthread_cond_signal(&td->free_cond);
761 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
764 if (!td_async_processing(td))
767 __sync_fetch_and_and(flags, ~value);
770 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
773 if (!td_async_processing(td))
776 __sync_fetch_and_or(flags, value);
779 extern const char *fio_get_arch_string(int);
780 extern const char *fio_get_os_string(int);
783 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
784 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
788 __FIO_OUTPUT_TERSE = 0,
789 __FIO_OUTPUT_JSON = 1,
790 __FIO_OUTPUT_NORMAL = 2,
791 __FIO_OUTPUT_JSON_PLUS = 3,
794 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
795 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
796 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
797 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
801 FIO_RAND_DIST_RANDOM = 0,
803 FIO_RAND_DIST_PARETO,
808 #define FIO_DEF_ZIPF 1.1
809 #define FIO_DEF_PARETO 0.2
812 FIO_RAND_GEN_TAUSWORTHE = 0,
814 FIO_RAND_GEN_TAUSWORTHE64,
822 extern void exec_trigger(const char *);
823 extern void check_trigger_file(void);