8 #include <sys/resource.h>
17 #include "compiler/compiler.h"
18 #include "thread_options.h"
21 #include "arch/arch.h"
35 #include "oslib/getopt.h"
37 #include "lib/rbtree.h"
42 #include "io_u_queue.h"
43 #include "workqueue.h"
45 #ifdef CONFIG_SOLARISAIO
46 #include <sys/asynch.h>
50 #include <linux/mempolicy.h>
54 * "local" is pseudo-policy
56 #define MPOL_LOCAL MPOL_MAX
60 * offset generator types
68 TD_F_VER_BACKLOG = 1U << 0,
69 TD_F_TRIM_BACKLOG = 1U << 1,
70 TD_F_READ_IOLOG = 1U << 2,
71 TD_F_REFILL_BUFFERS = 1U << 3,
72 TD_F_SCRAMBLE_BUFFERS = 1U << 4,
73 TD_F_VER_NONE = 1U << 5,
74 TD_F_PROFILE_OPS = 1U << 6,
75 TD_F_COMPRESS = 1U << 7,
77 TD_F_COMPRESS_LOG = 1U << 9,
78 TD_F_VSTATE_SAVED = 1U << 10,
79 TD_F_NEED_LOCK = 1U << 11,
80 TD_F_CHILD = 1U << 12,
81 TD_F_NO_PROGRESS = 1U << 13,
82 TD_F_REGROW_LOGS = 1U << 14,
91 FIO_RAND_FILE_SIZE_OFF,
94 FIO_RAND_SEQ_RAND_READ_OFF,
95 FIO_RAND_SEQ_RAND_WRITE_OFF,
96 FIO_RAND_SEQ_RAND_TRIM_OFF,
108 RATE_PROCESS_LINEAR = 0,
109 RATE_PROCESS_POISSON = 1,
113 * Per-thread/process specific data. Only used for the network client
117 void sk_out_assign(struct sk_out *);
118 void sk_out_drop(void);
120 struct zone_split_index {
122 uint8_t size_perc_prev;
126 * This describes a single thread/process executing a fio job.
129 struct thread_options o;
130 struct flist_head opt_list;
133 char verror[FIO_VERROR_SIZE];
135 unsigned int thread_number;
136 unsigned int subjob_number;
137 unsigned int groupid;
138 struct thread_stat ts;
142 struct io_log *slat_log;
143 struct io_log *clat_log;
144 struct io_log *lat_log;
145 struct io_log *bw_log;
146 struct io_log *iops_log;
148 struct workqueue log_compress_wq;
150 struct thread_data *parent;
152 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
153 struct timeval bw_sample_time;
155 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
156 struct timeval iops_sample_time;
158 volatile int update_rusage;
159 struct fio_mutex *rusage_sem;
160 struct rusage ru_start;
161 struct rusage ru_end;
163 struct fio_file **files;
164 unsigned char *file_locks;
165 unsigned int files_size;
166 unsigned int files_index;
167 unsigned int nr_open_files;
168 unsigned int nr_done_files;
169 unsigned int nr_normal_files;
171 unsigned int next_file;
172 struct frand_state next_file_state;
175 struct zipf_state next_file_zipf;
176 struct gauss_state next_file_gauss;
189 size_t orig_buffer_size;
190 volatile int terminate;
191 volatile int runstate;
192 unsigned int last_was_sync;
193 enum fio_ddir last_ddir;
202 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
204 struct frand_state bsrange_state;
205 struct frand_state verify_state;
206 struct frand_state trim_state;
207 struct frand_state delay_state;
209 struct frand_state buf_state;
210 struct frand_state buf_state_prev;
211 struct frand_state dedupe_state;
212 struct frand_state zone_state;
214 struct zone_split_index **zone_state_index;
216 unsigned int verify_batch;
217 unsigned int trim_batch;
219 struct thread_io_list *vstate;
224 * IO engine hooks, contains everything needed to submit an io_u
225 * to any of the available IO engines.
227 struct ioengine_ops *io_ops;
230 * Queue depth of io_u's that fio MIGHT do
232 unsigned int cur_depth;
235 * io_u's about to be committed
237 unsigned int io_u_queued;
240 * io_u's submitted but not completed yet
242 unsigned int io_u_in_flight;
245 * List of free and busy io_u's
247 struct io_u_ring io_u_requeues;
248 struct io_u_queue io_u_freelist;
249 struct io_u_queue io_u_all;
250 pthread_mutex_t io_u_lock;
251 pthread_cond_t free_cond;
254 * async verify offload
256 struct flist_head verify_list;
257 pthread_t *verify_threads;
258 unsigned int nr_verify_threads;
259 pthread_cond_t verify_cond;
260 int verify_thread_exit;
265 uint64_t rate_bps[DDIR_RWDIR_CNT];
266 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
267 unsigned long rate_bytes[DDIR_RWDIR_CNT];
268 unsigned long rate_blocks[DDIR_RWDIR_CNT];
269 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
270 struct timeval lastrate[DDIR_RWDIR_CNT];
272 struct frand_state poisson_state;
275 * Enforced rate submission/completion workqueue
277 struct workqueue io_wq;
279 uint64_t total_io_size;
280 uint64_t fill_device_size;
285 uint64_t io_issues[DDIR_RWDIR_CNT];
286 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
292 uint64_t io_blocks[DDIR_RWDIR_CNT];
293 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
294 uint64_t io_bytes[DDIR_RWDIR_CNT];
295 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
296 uint64_t io_skip_bytes;
298 struct fio_mutex *mutex;
299 uint64_t bytes_done[DDIR_RWDIR_CNT];
302 * State for random io, a bitmap of blocks done vs not done
304 struct frand_state random_state;
306 struct timeval start; /* start of this loop */
307 struct timeval epoch; /* time job was started */
308 struct timeval last_issue;
310 struct timeval tv_cache;
311 struct timeval terminate_time;
312 unsigned int tv_cache_nr;
313 unsigned int tv_cache_mask;
314 unsigned int ramp_time_over;
317 * Time since last latency_window was started
319 struct timeval latency_ts;
320 unsigned int latency_qd;
321 unsigned int latency_qd_high;
322 unsigned int latency_qd_low;
323 unsigned int latency_failed;
324 uint64_t latency_ios;
328 * read/write mixed workload state
330 struct frand_state rwmix_state;
331 unsigned long rwmix_issues;
332 enum fio_ddir rwmix_ddir;
333 unsigned int ddir_seq_nr;
336 * rand/seq mixed workload state
338 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
341 * IO history logs for verification. We use a tree for sorting,
342 * if we are overwriting. Otherwise just use a fifo.
344 struct rb_root io_hist_tree;
345 struct flist_head io_hist_list;
346 unsigned long io_hist_len;
351 struct flist_head io_log_list;
354 * For tracking/handling discards
356 struct flist_head trim_list;
357 unsigned long trim_entries;
359 struct flist_head next_rand_list;
362 * for fileservice, how often to switch to a new file
364 unsigned int file_service_nr;
365 unsigned int file_service_left;
366 struct fio_file *file_service_file;
368 unsigned int sync_file_range_nr;
371 * For generating file sizes
373 struct frand_state file_size_state;
378 unsigned int total_err_count;
381 struct fio_flow *flow;
384 * Can be overloaded by profiles
386 struct prof_io_ops prof_io_ops;
393 * when should interactive ETA output be generated
401 #define __td_verror(td, err, msg, func) \
403 unsigned int ____e = (err); \
406 (td)->error = ____e; \
407 if (!(td)->first_error) \
408 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
412 #define td_clear_error(td) do { \
415 (td)->parent->error = 0; \
418 #define td_verror(td, err, func) do { \
419 __td_verror((td), (err), strerror((err)), (func)); \
421 __td_verror((td)->parent, (err), strerror((err)), (func)); \
424 #define td_vmsg(td, err, msg, func) do { \
425 __td_verror((td), (err), (msg), (func)); \
427 __td_verror((td)->parent, (err), (msg), (func)); \
430 #define __fio_stringify_1(x) #x
431 #define __fio_stringify(x) __fio_stringify_1(x)
433 extern int exitall_on_terminate;
434 extern unsigned int thread_number;
435 extern unsigned int stat_number;
438 extern int output_format;
439 extern int append_terse_output;
440 extern int temp_stall_ts;
441 extern uintptr_t page_mask, page_size;
442 extern int read_only;
443 extern int eta_print;
444 extern int eta_new_line;
445 extern unsigned long done_secs;
446 extern char *job_section;
447 extern int fio_gtod_offload;
448 extern int fio_gtod_cpu;
449 extern enum fio_cs fio_clock_source;
450 extern int fio_clock_source_set;
451 extern int warnings_fatal;
452 extern int terse_version;
453 extern int is_backend;
454 extern int nr_clients;
455 extern int log_syslog;
456 extern int status_interval;
457 extern const char fio_version_string[];
458 extern char *trigger_file;
459 extern char *trigger_cmd;
460 extern char *trigger_remote_cmd;
461 extern long long trigger_timeout;
462 extern char *aux_path;
464 extern struct thread_data *threads;
466 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
468 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
471 #define REAL_MAX_JOBS 2048
473 static inline int should_fsync(struct thread_data *td)
475 if (td->last_was_sync)
477 if (td_write(td) || td_rw(td) || td->o.override_sync)
484 * Init/option functions
486 extern int __must_check fio_init_options(void);
487 extern int __must_check parse_options(int, char **);
488 extern int parse_jobs_ini(char *, int, int, int);
489 extern int parse_cmd_line(int, char **, int);
490 extern int fio_backend(struct sk_out *);
491 extern void reset_fio_state(void);
492 extern void clear_io_state(struct thread_data *, int);
493 extern int fio_options_parse(struct thread_data *, char **, int);
494 extern void fio_keywords_init(void);
495 extern void fio_keywords_exit(void);
496 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
497 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
498 extern void fio_fill_default_options(struct thread_data *);
499 extern int fio_show_option_help(const char *);
500 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
501 extern void fio_options_dup_and_init(struct option *);
502 extern void fio_options_mem_dupe(struct thread_data *);
503 extern void options_mem_dupe(void *data, struct fio_option *options);
504 extern void td_fill_rand_seeds(struct thread_data *);
505 extern void td_fill_verify_state_seed(struct thread_data *);
506 extern void add_job_opts(const char **, int);
507 extern char *num2str(uint64_t, int, int, int, int);
508 extern int ioengine_load(struct thread_data *);
509 extern int parse_dryrun(void);
510 extern int fio_running_or_pending_io_threads(void);
511 extern int fio_set_fd_nonblocking(int, const char *);
512 extern void sig_show_status(int sig);
513 extern struct thread_data *get_global_options(void);
515 extern uintptr_t page_mask;
516 extern uintptr_t page_size;
517 extern int initialize_fio(char *envp[]);
518 extern void deinitialize_fio(void);
520 #define FIO_GETOPT_JOB 0x89000000
521 #define FIO_GETOPT_IOENGINE 0x98000000
522 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
527 extern void print_thread_status(void);
528 extern void print_status_init(int);
529 extern char *fio_uint_to_kmg(unsigned int val);
532 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
533 * will never back again. It may cycle between running/verififying/fsyncing.
534 * Once the thread reaches TD_EXITED, it is just waiting for the core to
553 extern void td_set_runstate(struct thread_data *, int);
554 extern int td_bump_runstate(struct thread_data *, int);
555 extern void td_restore_runstate(struct thread_data *, int);
556 extern const char *runstate_to_name(int runstate);
559 * Allow 60 seconds for a job to quit on its own, otherwise reap with
562 #define FIO_REAP_TIMEOUT 60
564 #define TERMINATE_ALL (-1U)
565 extern void fio_terminate_threads(unsigned int);
566 extern void fio_mark_td_terminate(struct thread_data *);
571 extern int __must_check fio_pin_memory(struct thread_data *);
572 extern void fio_unpin_memory(struct thread_data *);
573 extern int __must_check allocate_io_mem(struct thread_data *);
574 extern void free_io_mem(struct thread_data *);
575 extern void free_threads_shm(void);
578 * Reset stats after ramp time completes
580 extern void reset_all_stats(struct thread_data *);
585 #ifdef FIO_HAVE_BLKTRACE
586 extern int is_blktrace(const char *, int *);
587 extern int load_blktrace(struct thread_data *, const char *, int);
590 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
591 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
592 struct timeval *comp_time);
595 * Latency target helpers
597 extern void lat_target_check(struct thread_data *);
598 extern void lat_target_init(struct thread_data *);
599 extern void lat_target_reset(struct thread_data *);
601 #define for_each_td(td, i) \
602 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
603 #define for_each_file(td, f, i) \
604 if ((td)->files_index) \
605 for ((i) = 0, (f) = (td)->files[0]; \
606 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
609 #define fio_assert(td, cond) do { \
612 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
613 td_set_runstate((td), TD_EXITED); \
614 (td)->error = EFAULT; \
619 static inline int fio_fill_issue_time(struct thread_data *td)
621 if (td->o.read_iolog_file ||
622 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
628 static inline int __should_check_rate(struct thread_data *td,
631 struct thread_options *o = &td->o;
634 * If some rate setting was given, we need to check it
636 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
637 o->rate_iops_min[ddir])
643 static inline int should_check_rate(struct thread_data *td)
647 if (td->bytes_done[DDIR_READ])
648 ret |= __should_check_rate(td, DDIR_READ);
649 if (td->bytes_done[DDIR_WRITE])
650 ret |= __should_check_rate(td, DDIR_WRITE);
651 if (td->bytes_done[DDIR_TRIM])
652 ret |= __should_check_rate(td, DDIR_TRIM);
657 static inline unsigned int td_max_bs(struct thread_data *td)
661 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
662 return max(td->o.max_bs[DDIR_TRIM], max_bs);
665 static inline unsigned int td_min_bs(struct thread_data *td)
669 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
670 return min(td->o.min_bs[DDIR_TRIM], min_bs);
673 static inline int td_async_processing(struct thread_data *td)
675 return (td->flags & TD_F_NEED_LOCK) != 0;
679 * We currently only need to do locking if we have verifier threads
680 * accessing our internal structures too
682 static inline void td_io_u_lock(struct thread_data *td)
684 if (td_async_processing(td))
685 pthread_mutex_lock(&td->io_u_lock);
688 static inline void td_io_u_unlock(struct thread_data *td)
690 if (td_async_processing(td))
691 pthread_mutex_unlock(&td->io_u_lock);
694 static inline void td_io_u_free_notify(struct thread_data *td)
696 if (td_async_processing(td))
697 pthread_cond_signal(&td->free_cond);
700 extern const char *fio_get_arch_string(int);
701 extern const char *fio_get_os_string(int);
704 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
705 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
709 __FIO_OUTPUT_TERSE = 0,
710 __FIO_OUTPUT_JSON = 1,
711 __FIO_OUTPUT_NORMAL = 2,
712 __FIO_OUTPUT_JSON_PLUS = 3,
715 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
716 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
717 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
718 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
722 FIO_RAND_DIST_RANDOM = 0,
724 FIO_RAND_DIST_PARETO,
729 #define FIO_DEF_ZIPF 1.1
730 #define FIO_DEF_PARETO 0.2
733 FIO_RAND_GEN_TAUSWORTHE = 0,
735 FIO_RAND_GEN_TAUSWORTHE64,
743 extern void exec_trigger(const char *);
744 extern void check_trigger_file(void);