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,
90 TD_F_MMAP_KEEP = 1U << 15,
101 FIO_RAND_FILE_SIZE_OFF,
104 FIO_RAND_SEQ_RAND_READ_OFF,
105 FIO_RAND_SEQ_RAND_WRITE_OFF,
106 FIO_RAND_SEQ_RAND_TRIM_OFF,
107 FIO_RAND_START_DELAY,
109 FIO_RAND_POISSON_OFF,
111 FIO_RAND_POISSON2_OFF,
112 FIO_RAND_POISSON3_OFF,
120 RATE_PROCESS_LINEAR = 0,
121 RATE_PROCESS_POISSON = 1,
132 * Per-thread/process specific data. Only used for the network client
135 void sk_out_assign(struct sk_out *);
136 void sk_out_drop(void);
138 struct zone_split_index {
140 uint8_t size_perc_prev;
144 * This describes a single thread/process executing a fio job.
147 struct flist_head opt_list;
149 struct thread_options o;
152 unsigned int thread_number;
153 unsigned int subjob_number;
154 unsigned int groupid;
155 struct thread_stat ts __attribute__ ((aligned(8)));
159 struct io_log *slat_log;
160 struct io_log *clat_log;
161 struct io_log *clat_hist_log;
162 struct io_log *lat_log;
163 struct io_log *bw_log;
164 struct io_log *iops_log;
166 struct workqueue log_compress_wq;
168 struct thread_data *parent;
170 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
171 struct timespec bw_sample_time;
173 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
174 struct timespec iops_sample_time;
176 volatile int update_rusage;
177 struct fio_mutex *rusage_sem;
178 struct rusage ru_start;
179 struct rusage ru_end;
181 struct fio_file **files;
182 unsigned char *file_locks;
183 unsigned int files_size;
184 unsigned int files_index;
185 unsigned int nr_open_files;
186 unsigned int nr_done_files;
187 unsigned int nr_normal_files;
189 unsigned int next_file;
190 struct frand_state next_file_state;
193 struct zipf_state next_file_zipf;
194 struct gauss_state next_file_gauss;
207 size_t orig_buffer_size;
208 volatile int terminate;
209 volatile int runstate;
210 unsigned int last_was_sync;
211 enum fio_ddir last_ddir;
218 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
220 struct frand_state bsrange_state[DDIR_RWDIR_CNT];
221 struct frand_state verify_state;
222 struct frand_state trim_state;
223 struct frand_state delay_state;
225 struct frand_state buf_state;
226 struct frand_state buf_state_prev;
227 struct frand_state dedupe_state;
228 struct frand_state zone_state;
230 struct zone_split_index **zone_state_index;
232 unsigned int verify_batch;
233 unsigned int trim_batch;
235 struct thread_io_list *vstate;
240 * IO engine hooks, contains everything needed to submit an io_u
241 * to any of the available IO engines.
243 struct ioengine_ops *io_ops;
247 * IO engine private data and dlhandle.
250 void *io_ops_dlhandle;
253 * Queue depth of io_u's that fio MIGHT do
255 unsigned int cur_depth;
258 * io_u's about to be committed
260 unsigned int io_u_queued;
263 * io_u's submitted but not completed yet
265 unsigned int io_u_in_flight;
268 * List of free and busy io_u's
270 struct io_u_ring io_u_requeues;
271 struct io_u_queue io_u_freelist;
272 struct io_u_queue io_u_all;
273 pthread_mutex_t io_u_lock;
274 pthread_cond_t free_cond;
277 * async verify offload
279 struct flist_head verify_list;
280 pthread_t *verify_threads;
281 unsigned int nr_verify_threads;
282 pthread_cond_t verify_cond;
283 int verify_thread_exit;
288 uint64_t rate_bps[DDIR_RWDIR_CNT];
289 uint64_t rate_next_io_time[DDIR_RWDIR_CNT];
290 unsigned long rate_bytes[DDIR_RWDIR_CNT];
291 unsigned long rate_blocks[DDIR_RWDIR_CNT];
292 unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
293 struct timespec lastrate[DDIR_RWDIR_CNT];
294 int64_t last_usec[DDIR_RWDIR_CNT];
295 struct frand_state poisson_state[DDIR_RWDIR_CNT];
298 * Enforced rate submission/completion workqueue
300 struct workqueue io_wq;
302 uint64_t total_io_size;
303 uint64_t fill_device_size;
308 uint64_t io_issues[DDIR_RWDIR_CNT];
309 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
315 uint64_t io_blocks[DDIR_RWDIR_CNT];
316 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
317 uint64_t io_bytes[DDIR_RWDIR_CNT];
318 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
319 uint64_t io_skip_bytes;
321 struct fio_mutex *mutex;
322 uint64_t bytes_done[DDIR_RWDIR_CNT];
325 * State for random io, a bitmap of blocks done vs not done
327 struct frand_state random_state;
329 struct timespec start; /* start of this loop */
330 struct timespec epoch; /* time job was started */
331 unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
332 struct timespec last_issue;
334 struct timespec ts_cache;
335 struct timespec terminate_time;
336 unsigned int ts_cache_nr;
337 unsigned int ts_cache_mask;
341 * Time since last latency_window was started
343 struct timespec latency_ts;
344 unsigned int latency_qd;
345 unsigned int latency_qd_high;
346 unsigned int latency_qd_low;
347 unsigned int latency_failed;
348 uint64_t latency_ios;
352 * read/write mixed workload state
354 struct frand_state rwmix_state;
355 unsigned long rwmix_issues;
356 enum fio_ddir rwmix_ddir;
357 unsigned int ddir_seq_nr;
360 * rand/seq mixed workload state
362 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
365 * IO history logs for verification. We use a tree for sorting,
366 * if we are overwriting. Otherwise just use a fifo.
368 struct rb_root io_hist_tree;
369 struct flist_head io_hist_list;
370 unsigned long io_hist_len;
375 struct flist_head io_log_list;
378 * For tracking/handling discards
380 struct flist_head trim_list;
381 unsigned long trim_entries;
383 struct flist_head next_rand_list;
386 * for fileservice, how often to switch to a new file
388 unsigned int file_service_nr;
389 unsigned int file_service_left;
390 struct fio_file *file_service_file;
392 unsigned int sync_file_range_nr;
395 * For generating file sizes
397 struct frand_state file_size_state;
402 unsigned int total_err_count;
405 struct fio_flow *flow;
408 * Can be overloaded by profiles
410 struct prof_io_ops prof_io_ops;
415 struct steadystate_data ss;
417 char verror[FIO_VERROR_SIZE];
421 * for GPU memory management
427 CUdeviceptr dev_mem_ptr;
433 * when should interactive ETA output be generated
441 #define __td_verror(td, err, msg, func) \
443 unsigned int ____e = (err); \
446 (td)->error = ____e; \
447 if (!(td)->first_error) \
448 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
452 #define td_clear_error(td) do { \
455 (td)->parent->error = 0; \
458 #define td_verror(td, err, func) do { \
459 __td_verror((td), (err), strerror((err)), (func)); \
461 __td_verror((td)->parent, (err), strerror((err)), (func)); \
464 #define td_vmsg(td, err, msg, func) do { \
465 __td_verror((td), (err), (msg), (func)); \
467 __td_verror((td)->parent, (err), (msg), (func)); \
470 #define __fio_stringify_1(x) #x
471 #define __fio_stringify(x) __fio_stringify_1(x)
473 extern int exitall_on_terminate;
474 extern unsigned int thread_number;
475 extern unsigned int stat_number;
478 extern int output_format;
479 extern int append_terse_output;
480 extern int temp_stall_ts;
481 extern uintptr_t page_mask, page_size;
482 extern int read_only;
483 extern int eta_print;
484 extern int eta_new_line;
485 extern unsigned long done_secs;
486 extern int fio_gtod_offload;
487 extern int fio_gtod_cpu;
488 extern enum fio_cs fio_clock_source;
489 extern int fio_clock_source_set;
490 extern int warnings_fatal;
491 extern int terse_version;
492 extern int is_backend;
493 extern int nr_clients;
494 extern int log_syslog;
495 extern int status_interval;
496 extern const char fio_version_string[];
497 extern char *trigger_file;
498 extern char *trigger_cmd;
499 extern char *trigger_remote_cmd;
500 extern long long trigger_timeout;
501 extern char *aux_path;
503 extern struct thread_data *threads;
505 static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
507 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
510 #define REAL_MAX_JOBS 4096
512 static inline int should_fsync(struct thread_data *td)
514 if (td->last_was_sync)
516 if (td_write(td) || td->o.override_sync)
523 * Init/option functions
525 extern int __must_check fio_init_options(void);
526 extern int __must_check parse_options(int, char **);
527 extern int parse_jobs_ini(char *, int, int, int);
528 extern int parse_cmd_line(int, char **, int);
529 extern int fio_backend(struct sk_out *);
530 extern void reset_fio_state(void);
531 extern void clear_io_state(struct thread_data *, int);
532 extern int fio_options_parse(struct thread_data *, char **, int);
533 extern void fio_keywords_init(void);
534 extern void fio_keywords_exit(void);
535 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
536 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
537 extern void fio_fill_default_options(struct thread_data *);
538 extern int fio_show_option_help(const char *);
539 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
540 extern void fio_options_dup_and_init(struct option *);
541 extern void fio_options_mem_dupe(struct thread_data *);
542 extern void td_fill_rand_seeds(struct thread_data *);
543 extern void td_fill_verify_state_seed(struct thread_data *);
544 extern void add_job_opts(const char **, int);
545 extern int ioengine_load(struct thread_data *);
546 extern bool parse_dryrun(void);
547 extern int fio_running_or_pending_io_threads(void);
548 extern int fio_set_fd_nonblocking(int, const char *);
549 extern void sig_show_status(int sig);
550 extern struct thread_data *get_global_options(void);
552 extern uintptr_t page_mask;
553 extern uintptr_t page_size;
554 extern int initialize_fio(char *envp[]);
555 extern void deinitialize_fio(void);
557 #define FIO_GETOPT_JOB 0x89000000
558 #define FIO_GETOPT_IOENGINE 0x98000000
559 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
564 extern void print_thread_status(void);
565 extern void print_status_init(int);
566 extern char *fio_uint_to_kmg(unsigned int val);
569 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
570 * will never back again. It may cycle between running/verififying/fsyncing.
571 * Once the thread reaches TD_EXITED, it is just waiting for the core to
591 #define TD_ENG_FLAG_SHIFT 16
592 #define TD_ENG_FLAG_MASK ((1U << 16) - 1)
594 static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
596 return (enum fio_ioengine_flags)
597 ((td->flags >> TD_ENG_FLAG_SHIFT) & TD_ENG_FLAG_MASK);
600 static inline void td_set_ioengine_flags(struct thread_data *td)
602 td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
603 (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
606 static inline bool td_ioengine_flagged(struct thread_data *td,
607 enum fio_ioengine_flags flags)
609 return ((td->flags >> TD_ENG_FLAG_SHIFT) & flags) != 0;
612 extern void td_set_runstate(struct thread_data *, int);
613 extern int td_bump_runstate(struct thread_data *, int);
614 extern void td_restore_runstate(struct thread_data *, int);
615 extern const char *runstate_to_name(int runstate);
618 * Allow 60 seconds for a job to quit on its own, otherwise reap with
621 #define FIO_REAP_TIMEOUT 300
623 #define TERMINATE_ALL (-1U)
624 extern void fio_terminate_threads(unsigned int);
625 extern void fio_mark_td_terminate(struct thread_data *);
630 extern int __must_check fio_pin_memory(struct thread_data *);
631 extern void fio_unpin_memory(struct thread_data *);
632 extern int __must_check allocate_io_mem(struct thread_data *);
633 extern void free_io_mem(struct thread_data *);
634 extern void free_threads_shm(void);
637 #define PTR_ALIGN(ptr, mask) \
638 (char *) (((uintptr_t) (ptr) + (mask)) & ~(mask))
642 * Reset stats after ramp time completes
644 extern void reset_all_stats(struct thread_data *);
646 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
647 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
648 struct timespec *comp_time);
651 * Latency target helpers
653 extern void lat_target_check(struct thread_data *);
654 extern void lat_target_init(struct thread_data *);
655 extern void lat_target_reset(struct thread_data *);
658 * Iterates all threads/processes within all the defined jobs
660 #define for_each_td(td, i) \
661 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
662 #define for_each_file(td, f, i) \
663 if ((td)->files_index) \
664 for ((i) = 0, (f) = (td)->files[0]; \
665 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
668 #define fio_assert(td, cond) do { \
671 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
672 td_set_runstate((td), TD_EXITED); \
673 (td)->error = EFAULT; \
678 static inline bool fio_fill_issue_time(struct thread_data *td)
680 if (td->o.read_iolog_file ||
681 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
687 static inline bool __should_check_rate(struct thread_data *td,
690 struct thread_options *o = &td->o;
693 * If some rate setting was given, we need to check it
695 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
696 o->rate_iops_min[ddir])
702 static inline bool should_check_rate(struct thread_data *td)
704 if (td->bytes_done[DDIR_READ] && __should_check_rate(td, DDIR_READ))
706 if (td->bytes_done[DDIR_WRITE] && __should_check_rate(td, DDIR_WRITE))
708 if (td->bytes_done[DDIR_TRIM] && __should_check_rate(td, DDIR_TRIM))
714 static inline unsigned int td_max_bs(struct thread_data *td)
718 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
719 return max(td->o.max_bs[DDIR_TRIM], max_bs);
722 static inline unsigned int td_min_bs(struct thread_data *td)
726 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
727 return min(td->o.min_bs[DDIR_TRIM], min_bs);
730 static inline bool td_async_processing(struct thread_data *td)
732 return (td->flags & TD_F_NEED_LOCK) != 0;
736 * We currently only need to do locking if we have verifier threads
737 * accessing our internal structures too
739 static inline void td_io_u_lock(struct thread_data *td)
741 if (td_async_processing(td))
742 pthread_mutex_lock(&td->io_u_lock);
745 static inline void td_io_u_unlock(struct thread_data *td)
747 if (td_async_processing(td))
748 pthread_mutex_unlock(&td->io_u_lock);
751 static inline void td_io_u_free_notify(struct thread_data *td)
753 if (td_async_processing(td))
754 pthread_cond_signal(&td->free_cond);
757 static inline void td_flags_clear(struct thread_data *td, unsigned int *flags,
760 if (!td_async_processing(td))
763 __sync_fetch_and_and(flags, ~value);
766 static inline void td_flags_set(struct thread_data *td, unsigned int *flags,
769 if (!td_async_processing(td))
772 __sync_fetch_and_or(flags, value);
775 extern const char *fio_get_arch_string(int);
776 extern const char *fio_get_os_string(int);
779 #define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
780 #define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
784 __FIO_OUTPUT_TERSE = 0,
785 __FIO_OUTPUT_JSON = 1,
786 __FIO_OUTPUT_NORMAL = 2,
787 __FIO_OUTPUT_JSON_PLUS = 3,
790 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
791 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
792 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
793 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
797 FIO_RAND_DIST_RANDOM = 0,
799 FIO_RAND_DIST_PARETO,
804 #define FIO_DEF_ZIPF 1.1
805 #define FIO_DEF_PARETO 0.2
808 FIO_RAND_GEN_TAUSWORTHE = 0,
810 FIO_RAND_GEN_TAUSWORTHE64,
818 extern void exec_trigger(const char *);
819 extern void check_trigger_file(void);