8 #include <sys/resource.h>
19 #include "compiler/compiler.h"
23 #include "arch/arch.h"
37 #include "lib/getopt.h"
47 #ifdef FIO_HAVE_SOLARISAIO
48 #include <sys/asynch.h>
52 * What type of allocation to use for io buffers
55 MEM_MALLOC = 0, /* ordinary malloc */
56 MEM_SHM, /* use shared memory segments */
57 MEM_SHMHUGE, /* use shared memory segments with huge pages */
58 MEM_MMAP, /* use anonynomous mmap */
59 MEM_MMAPHUGE, /* memory mapped huge file */
63 * offset generator types
71 * What type of errors to continue on when continue_on_error is used
74 ERROR_TYPE_READ_BIT = 0,
75 ERROR_TYPE_WRITE_BIT = 1,
76 ERROR_TYPE_VERIFY_BIT = 2,
82 ERROR_TYPE_READ = 1 << ERROR_TYPE_READ_BIT,
83 ERROR_TYPE_WRITE = 1 << ERROR_TYPE_WRITE_BIT,
84 ERROR_TYPE_VERIFY = 1 << ERROR_TYPE_VERIFY_BIT,
85 ERROR_TYPE_ANY = 0xffff,
93 struct thread_options {
101 enum td_ddir td_ddir;
103 unsigned int kb_base;
104 unsigned int ddir_seq_nr;
106 unsigned int iodepth;
107 unsigned int iodepth_low;
108 unsigned int iodepth_batch;
109 unsigned int iodepth_batch_complete;
111 unsigned long long size;
112 unsigned int size_percent;
113 unsigned int fill_device;
114 unsigned long long file_size_low;
115 unsigned long long file_size_high;
116 unsigned long long start_offset;
118 unsigned int bs[DDIR_RWDIR_CNT];
119 unsigned int ba[DDIR_RWDIR_CNT];
120 unsigned int min_bs[DDIR_RWDIR_CNT];
121 unsigned int max_bs[DDIR_RWDIR_CNT];
122 struct bssplit *bssplit[DDIR_RWDIR_CNT];
123 unsigned int bssplit_nr[DDIR_RWDIR_CNT];
125 int *ignore_error[ERROR_TYPE_CNT];
126 unsigned int ignore_error_nr[ERROR_TYPE_CNT];
127 unsigned int error_dump;
129 unsigned int nr_files;
130 unsigned int open_files;
131 enum file_lock_mode file_lock_mode;
132 unsigned int lockfile_batch;
134 unsigned int odirect;
135 unsigned int invalidate_cache;
136 unsigned int create_serialize;
137 unsigned int create_fsync;
138 unsigned int create_on_open;
139 unsigned int create_only;
140 unsigned int end_fsync;
141 unsigned int pre_read;
142 unsigned int sync_io;
144 unsigned int do_verify;
145 unsigned int verifysort;
146 unsigned int verify_interval;
147 unsigned int verify_offset;
148 char verify_pattern[MAX_PATTERN_SIZE];
149 unsigned int verify_pattern_bytes;
150 unsigned int verify_fatal;
151 unsigned int verify_dump;
152 unsigned int verify_async;
153 unsigned long long verify_backlog;
154 unsigned int verify_batch;
155 unsigned int use_thread;
157 unsigned int do_disk_util;
158 unsigned int override_sync;
159 unsigned int rand_repeatable;
160 unsigned int use_os_rand;
161 unsigned int write_lat_log;
162 unsigned int write_bw_log;
163 unsigned int write_iops_log;
164 unsigned int log_avg_msec;
165 unsigned int norandommap;
166 unsigned int softrandommap;
167 unsigned int bs_unaligned;
168 unsigned int fsync_on_close;
170 unsigned int hugepage_size;
171 unsigned int rw_min_bs;
172 unsigned int thinktime;
173 unsigned int thinktime_spin;
174 unsigned int thinktime_blocks;
175 unsigned int fsync_blocks;
176 unsigned int fdatasync_blocks;
177 unsigned int barrier_blocks;
178 unsigned long long start_delay;
179 unsigned long long timeout;
180 unsigned long long ramp_time;
181 unsigned int overwrite;
182 unsigned int bw_avg_time;
183 unsigned int iops_avg_time;
185 unsigned long long zone_range;
186 unsigned long long zone_size;
187 unsigned long long zone_skip;
188 enum fio_memtype mem_type;
189 unsigned int mem_align;
191 unsigned int stonewall;
192 unsigned int new_group;
193 unsigned int numjobs;
194 os_cpu_mask_t cpumask;
195 unsigned int cpumask_set;
196 os_cpu_mask_t verify_cpumask;
197 unsigned int verify_cpumask_set;
199 unsigned int rwmixcycle;
200 unsigned int rwmix[2];
202 unsigned int file_service_type;
203 unsigned int group_reporting;
204 unsigned int fadvise_hint;
205 enum fio_fallocate_mode fallocate_mode;
206 unsigned int zero_buffers;
207 unsigned int refill_buffers;
208 unsigned int scramble_buffers;
209 unsigned int compress_percentage;
210 unsigned int compress_chunk;
211 unsigned int time_based;
212 unsigned int disable_lat;
213 unsigned int disable_clat;
214 unsigned int disable_slat;
215 unsigned int disable_bw;
216 unsigned int gtod_reduce;
217 unsigned int gtod_cpu;
218 unsigned int gtod_offload;
219 enum fio_cs clocksource;
220 unsigned int no_stall;
221 unsigned int trim_percentage;
222 unsigned int trim_batch;
223 unsigned int trim_zero;
224 unsigned long long trim_backlog;
225 unsigned int clat_percentiles;
226 unsigned int overwrite_plist;
227 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
229 char *read_iolog_file;
230 char *write_iolog_file;
234 char *replay_redirect;
237 * Pre-run and post-run shell
242 unsigned int rate[DDIR_RWDIR_CNT];
243 unsigned int ratemin[DDIR_RWDIR_CNT];
244 unsigned int ratecycle;
245 unsigned int rate_iops[DDIR_RWDIR_CNT];
246 unsigned int rate_iops_min[DDIR_RWDIR_CNT];
251 * CPU "io" cycle burner
253 unsigned int cpuload;
254 unsigned int cpucycle;
259 enum error_type continue_on_error;
262 * Benchmark profile type
267 * blkio cgroup support
270 unsigned int cgroup_weight;
271 unsigned int cgroup_nodelete;
279 unsigned int flow_sleep;
281 unsigned long long offset_increment;
283 unsigned int sync_file_range;
287 * This describes a single thread/process executing a fio job.
290 struct thread_options o;
292 char verror[FIO_VERROR_SIZE];
296 struct thread_stat ts;
298 struct io_log *slat_log;
299 struct io_log *clat_log;
300 struct io_log *lat_log;
301 struct io_log *bw_log;
302 struct io_log *iops_log;
304 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
305 struct timeval bw_sample_time;
307 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
308 struct timeval iops_sample_time;
310 struct rusage ru_start;
311 struct rusage ru_end;
313 struct fio_file **files;
314 unsigned int files_size;
315 unsigned int files_index;
316 unsigned int nr_open_files;
317 unsigned int nr_done_files;
318 unsigned int nr_normal_files;
320 unsigned int next_file;
321 os_random_state_t next_file_state;
322 struct frand_state __next_file_state;
329 size_t orig_buffer_size;
330 volatile int terminate;
331 volatile int runstate;
333 unsigned int ioprio_set;
334 unsigned int last_was_sync;
335 enum fio_ddir last_ddir;
345 unsigned long rand_seeds[8];
348 os_random_state_t bsrange_state;
349 struct frand_state __bsrange_state;
352 os_random_state_t verify_state;
353 struct frand_state __verify_state;
356 os_random_state_t trim_state;
357 struct frand_state __trim_state;
360 struct frand_state buf_state;
362 unsigned int verify_batch;
363 unsigned int trim_batch;
368 * IO engine hooks, contains everything needed to submit an io_u
369 * to any of the available IO engines.
371 struct ioengine_ops *io_ops;
374 * Queue depth of io_u's that fio MIGHT do
376 unsigned int cur_depth;
379 * io_u's about to be committed
381 unsigned int io_u_queued;
384 * io_u's submitted but not completed yet
386 unsigned int io_u_in_flight;
389 * List of free and busy io_u's
391 struct flist_head io_u_freelist;
392 struct flist_head io_u_busylist;
393 struct flist_head io_u_requeues;
394 pthread_mutex_t io_u_lock;
395 pthread_cond_t free_cond;
398 * async verify offload
400 struct flist_head verify_list;
401 pthread_t *verify_threads;
402 unsigned int nr_verify_threads;
403 pthread_cond_t verify_cond;
404 int verify_thread_exit;
409 unsigned long long rate_bps[DDIR_RWDIR_CNT];
410 long rate_pending_usleep[DDIR_RWDIR_CNT];
411 unsigned long rate_bytes[DDIR_RWDIR_CNT];
412 unsigned long rate_blocks[DDIR_RWDIR_CNT];
413 struct timeval lastrate[DDIR_RWDIR_CNT];
415 unsigned long long total_io_size;
416 unsigned long long fill_device_size;
418 unsigned long io_issues[DDIR_RWDIR_CNT];
419 unsigned long long io_blocks[DDIR_RWDIR_CNT];
420 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
421 unsigned long long io_bytes[DDIR_RWDIR_CNT];
422 unsigned long long io_skip_bytes;
423 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
424 unsigned long long zone_bytes;
425 struct fio_mutex *mutex;
428 * State for random io, a bitmap of blocks done vs not done
431 os_random_state_t random_state;
432 struct frand_state __random_state;
435 struct timeval start; /* start of this loop */
436 struct timeval epoch; /* time job was started */
437 struct timeval last_issue;
438 struct timeval tv_cache;
439 unsigned int tv_cache_nr;
440 unsigned int tv_cache_mask;
441 unsigned int ramp_time_over;
444 * read/write mixed workload state
447 os_random_state_t rwmix_state;
448 struct frand_state __rwmix_state;
450 unsigned long rwmix_issues;
451 enum fio_ddir rwmix_ddir;
452 unsigned int ddir_seq_nr;
455 * IO history logs for verification. We use a tree for sorting,
456 * if we are overwriting. Otherwise just use a fifo.
458 struct rb_root io_hist_tree;
459 struct flist_head io_hist_list;
460 unsigned long io_hist_len;
465 struct flist_head io_log_list;
468 * For tracking/handling discards
470 struct flist_head trim_list;
471 unsigned long trim_entries;
474 * for fileservice, how often to switch to a new file
476 unsigned int file_service_nr;
477 unsigned int file_service_left;
478 struct fio_file *file_service_file;
480 unsigned int sync_file_range_nr;
483 * For generating file sizes
486 os_random_state_t file_size_state;
487 struct frand_state __file_size_state;
493 unsigned int total_err_count;
496 struct fio_flow *flow;
499 * Can be overloaded by profiles
501 struct prof_io_ops prof_io_ops;
506 * when should interactive ETA output be generated
514 #define __td_verror(td, err, msg, func) \
520 if (!(td)->first_error) \
521 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
525 #define td_clear_error(td) \
527 #define td_verror(td, err, func) \
528 __td_verror((td), (err), strerror((err)), (func))
529 #define td_vmsg(td, err, msg, func) \
530 __td_verror((td), (err), (msg), (func))
532 #define __fio_stringify_1(x) #x
533 #define __fio_stringify(x) __fio_stringify_1(x)
535 extern int exitall_on_terminate;
536 extern unsigned int thread_number;
537 extern unsigned int nr_process, nr_thread;
540 extern int output_format;
541 extern int temp_stall_ts;
542 extern unsigned long long mlock_size;
543 extern uintptr_t page_mask, page_size;
544 extern int read_only;
545 extern int eta_print;
546 extern unsigned long done_secs;
547 extern char *job_section;
548 extern int fio_gtod_offload;
549 extern int fio_gtod_cpu;
550 extern enum fio_cs fio_clock_source;
551 extern int warnings_fatal;
552 extern int terse_version;
553 extern int is_backend;
554 extern int nr_clients;
555 extern int log_syslog;
556 extern const char fio_version_string[];
557 extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
559 extern struct thread_data *threads;
561 static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
563 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
566 #define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
567 #define TO_MAP_BLOCK(f, b) (b)
568 #define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
569 #define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
571 #define REAL_MAX_JOBS 2048
573 static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
576 return ERROR_TYPE_VERIFY_BIT;
577 if (ddir == DDIR_READ)
578 return ERROR_TYPE_READ_BIT;
579 return ERROR_TYPE_WRITE_BIT;
582 static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
583 static inline int td_non_fatal_error(struct thread_data *td,
584 enum error_type_bit etype, int err)
587 if (!td->o.ignore_error[etype]) {
588 td->o.ignore_error[etype] = __NON_FATAL_ERR;
589 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
593 if (!(td->o.continue_on_error & (1 << etype)))
595 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
596 if (td->o.ignore_error[etype][i] == err)
601 static inline void update_error_count(struct thread_data *td, int err)
603 td->total_err_count++;
604 if (td->total_err_count == 1)
605 td->first_error = err;
608 static inline int should_fsync(struct thread_data *td)
610 if (td->last_was_sync)
614 if (td_write(td) || td_rw(td) || td->o.override_sync)
621 * Init/option functions
623 extern int __must_check parse_options(int, char **);
624 extern int parse_jobs_ini(char *, int, int);
625 extern int parse_cmd_line(int, char **);
626 extern int fio_backend(void);
627 extern void reset_fio_state(void);
628 extern void clear_io_state(struct thread_data *);
629 extern int fio_options_parse(struct thread_data *, char **, int);
630 extern void fio_keywords_init(void);
631 extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
632 extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
633 extern void fio_fill_default_options(struct thread_data *);
634 extern int fio_show_option_help(const char *);
635 extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
636 extern void fio_options_dup_and_init(struct option *);
637 extern void fio_options_mem_dupe(struct thread_data *);
638 extern void options_mem_dupe(void *data, struct fio_option *options);
639 extern void td_fill_rand_seeds(struct thread_data *);
640 extern void add_job_opts(const char **);
641 extern char *num2str(unsigned long, int, int, int);
642 extern int ioengine_load(struct thread_data *);
644 #define FIO_GETOPT_JOB 0x89000000
645 #define FIO_GETOPT_IOENGINE 0x98000000
646 #define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
651 extern void print_thread_status(void);
652 extern void print_status_init(int);
655 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
656 * will never back again. It may cycle between running/verififying/fsyncing.
657 * Once the thread reaches TD_EXITED, it is just waiting for the core to
673 extern void td_set_runstate(struct thread_data *, int);
674 #define TERMINATE_ALL (-1)
675 extern void fio_terminate_threads(int);
680 extern int __must_check fio_pin_memory(void);
681 extern void fio_unpin_memory(void);
682 extern int __must_check allocate_io_mem(struct thread_data *);
683 extern void free_io_mem(struct thread_data *);
686 * Reset stats after ramp time completes
688 extern void reset_all_stats(struct thread_data *);
693 #ifdef FIO_HAVE_BLKTRACE
694 extern int is_blktrace(const char *);
695 extern int load_blktrace(struct thread_data *, const char *);
699 * Mark unused variables passed to ops functions as unused, to silence gcc
701 #define fio_unused __attribute((__unused__))
702 #define fio_init __attribute__((constructor))
703 #define fio_exit __attribute__((destructor))
705 #define for_each_td(td, i) \
706 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
707 #define for_each_file(td, f, i) \
708 if ((td)->files_index) \
709 for ((i) = 0, (f) = (td)->files[0]; \
710 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
713 #define fio_assert(td, cond) do { \
716 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
717 td_set_runstate((td), TD_EXITED); \
718 (td)->error = EFAULT; \
723 static inline int fio_fill_issue_time(struct thread_data *td)
725 if (td->o.read_iolog_file ||
726 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
732 static inline int __should_check_rate(struct thread_data *td,
735 struct thread_options *o = &td->o;
738 * If some rate setting was given, we need to check it
740 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
741 o->rate_iops_min[ddir])
747 static inline int should_check_rate(struct thread_data *td,
748 unsigned long *bytes_done)
752 if (bytes_done[DDIR_READ])
753 ret |= __should_check_rate(td, DDIR_READ);
754 if (bytes_done[DDIR_WRITE])
755 ret |= __should_check_rate(td, DDIR_WRITE);
756 if (bytes_done[DDIR_TRIM])
757 ret |= __should_check_rate(td, DDIR_TRIM);
762 static inline int is_power_of_2(unsigned int val)
764 return (val != 0 && ((val & (val - 1)) == 0));
768 * We currently only need to do locking if we have verifier threads
769 * accessing our internal structures too
771 static inline void td_io_u_lock(struct thread_data *td)
773 if (td->o.verify_async)
774 pthread_mutex_lock(&td->io_u_lock);
777 static inline void td_io_u_unlock(struct thread_data *td)
779 if (td->o.verify_async)
780 pthread_mutex_unlock(&td->io_u_lock);
783 static inline void td_io_u_free_notify(struct thread_data *td)
785 if (td->o.verify_async)
786 pthread_cond_signal(&td->free_cond);
789 extern const char *fio_get_arch_string(int);
790 extern const char *fio_get_os_string(int);
793 FIO_OUTPUT_TERSE = 0,