projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
t/btrace2fio: don't dec inflight for requeue
[fio.git] / fio.h
... / ...
CommitLineData
1#ifndef FIO_H
2#define FIO_H
3
4#include <sched.h>
5#include <limits.h>
6#include <pthread.h>
7#include <sys/time.h>
8#include <sys/resource.h>
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
12#include <unistd.h>
13#include <string.h>
14#include <inttypes.h>
15#include <assert.h>
16
17#include "compiler/compiler.h"
18#include "thread_options.h"
19#include "flist.h"
20#include "fifo.h"
21#include "arch/arch.h"
22#include "os/os.h"
23#include "mutex.h"
24#include "log.h"
25#include "debug.h"
26#include "file.h"
27#include "io_ddir.h"
28#include "ioengine.h"
29#include "iolog.h"
30#include "helpers.h"
31#include "options.h"
32#include "profile.h"
33#include "fio_time.h"
34#include "gettime.h"
35#include "lib/getopt.h"
36#include "lib/rand.h"
37#include "lib/rbtree.h"
38#include "client.h"
39#include "server.h"
40#include "stat.h"
41#include "flow.h"
42#include "io_u_queue.h"
43
44#ifdef CONFIG_SOLARISAIO
45#include <sys/asynch.h>
46#endif
47
48#ifdef CONFIG_LIBNUMA
49#include <linux/mempolicy.h>
50#include <numa.h>
51
52/*
53 * "local" is pseudo-policy
54 */
55#define MPOL_LOCAL MPOL_MAX
56#endif
57
58/*
59 * offset generator types
60 */
61enum {
62 RW_SEQ_SEQ = 0,
63 RW_SEQ_IDENT,
64};
65
66enum {
67 TD_F_VER_BACKLOG = 1,
68 TD_F_TRIM_BACKLOG = 2,
69 TD_F_READ_IOLOG = 4,
70 TD_F_REFILL_BUFFERS = 8,
71 TD_F_SCRAMBLE_BUFFERS = 16,
72 TD_F_VER_NONE = 32,
73 TD_F_PROFILE_OPS = 64,
74 TD_F_COMPRESS = 128,
75 TD_F_NOIO = 256,
76 TD_F_COMPRESS_LOG = 512,
77};
78
79enum {
80 FIO_RAND_BS_OFF = 0,
81 FIO_RAND_VER_OFF,
82 FIO_RAND_MIX_OFF,
83 FIO_RAND_FILE_OFF,
84 FIO_RAND_BLOCK_OFF,
85 FIO_RAND_FILE_SIZE_OFF,
86 FIO_RAND_TRIM_OFF,
87 FIO_RAND_BUF_OFF,
88 FIO_RAND_SEQ_RAND_READ_OFF,
89 FIO_RAND_SEQ_RAND_WRITE_OFF,
90 FIO_RAND_SEQ_RAND_TRIM_OFF,
91 FIO_RAND_START_DELAY,
92 FIO_DEDUPE_OFF,
93 FIO_RAND_NR_OFFS,
94};
95
96/*
97 * This describes a single thread/process executing a fio job.
98 */
99struct thread_data {
100 struct thread_options o;
101 unsigned long flags;
102 void *eo;
103 char verror[FIO_VERROR_SIZE];
104 pthread_t thread;
105 unsigned int thread_number;
106 unsigned int subjob_number;
107 unsigned int groupid;
108 struct thread_stat ts;
109
110 int client_type;
111
112 struct io_log *slat_log;
113 struct io_log *clat_log;
114 struct io_log *lat_log;
115 struct io_log *bw_log;
116 struct io_log *iops_log;
117
118 struct tp_data *tp_data;
119
120 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
121 struct timeval bw_sample_time;
122
123 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
124 struct timeval iops_sample_time;
125
126 volatile int update_rusage;
127 struct fio_mutex *rusage_sem;
128 struct rusage ru_start;
129 struct rusage ru_end;
130
131 struct fio_file **files;
132 unsigned char *file_locks;
133 unsigned int files_size;
134 unsigned int files_index;
135 unsigned int nr_open_files;
136 unsigned int nr_done_files;
137 unsigned int nr_normal_files;
138 union {
139 unsigned int next_file;
140 struct frand_state next_file_state;
141 };
142 int error;
143 int sig;
144 int done;
145 pid_t pid;
146 char *orig_buffer;
147 size_t orig_buffer_size;
148 volatile int terminate;
149 volatile int runstate;
150 unsigned int last_was_sync;
151 enum fio_ddir last_ddir;
152
153 int mmapfd;
154
155 void *iolog_buf;
156 FILE *iolog_f;
157
158 char *sysfs_root;
159
160 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
161
162 struct frand_state bsrange_state;
163 struct frand_state verify_state;
164 struct frand_state trim_state;
165 struct frand_state delay_state;
166
167 struct frand_state buf_state;
168 struct frand_state buf_state_prev;
169 struct frand_state dedupe_state;
170
171 unsigned int verify_batch;
172 unsigned int trim_batch;
173
174 int shm_id;
175
176 /*
177 * IO engine hooks, contains everything needed to submit an io_u
178 * to any of the available IO engines.
179 */
180 struct ioengine_ops *io_ops;
181
182 /*
183 * Queue depth of io_u's that fio MIGHT do
184 */
185 unsigned int cur_depth;
186
187 /*
188 * io_u's about to be committed
189 */
190 unsigned int io_u_queued;
191
192 /*
193 * io_u's submitted but not completed yet
194 */
195 unsigned int io_u_in_flight;
196
197 /*
198 * List of free and busy io_u's
199 */
200 struct io_u_ring io_u_requeues;
201 struct io_u_queue io_u_freelist;
202 struct io_u_queue io_u_all;
203 pthread_mutex_t io_u_lock;
204 pthread_cond_t free_cond;
205
206 /*
207 * async verify offload
208 */
209 struct flist_head verify_list;
210 pthread_t *verify_threads;
211 unsigned int nr_verify_threads;
212 pthread_cond_t verify_cond;
213 int verify_thread_exit;
214
215 /*
216 * Rate state
217 */
218 uint64_t rate_bps[DDIR_RWDIR_CNT];
219 long rate_pending_usleep[DDIR_RWDIR_CNT];
220 unsigned long rate_bytes[DDIR_RWDIR_CNT];
221 unsigned long rate_blocks[DDIR_RWDIR_CNT];
222 struct timeval lastrate[DDIR_RWDIR_CNT];
223
224 uint64_t total_io_size;
225 uint64_t fill_device_size;
226
227 unsigned long io_issues[DDIR_RWDIR_CNT];
228 uint64_t io_blocks[DDIR_RWDIR_CNT];
229 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
230 uint64_t io_bytes[DDIR_RWDIR_CNT];
231 uint64_t io_skip_bytes;
232 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
233 uint64_t zone_bytes;
234 struct fio_mutex *mutex;
235
236 /*
237 * State for random io, a bitmap of blocks done vs not done
238 */
239 struct frand_state random_state;
240
241 struct timeval start; /* start of this loop */
242 struct timeval epoch; /* time job was started */
243 struct timeval last_issue;
244 struct timeval tv_cache;
245 struct timeval terminate_time;
246 unsigned int tv_cache_nr;
247 unsigned int tv_cache_mask;
248 unsigned int ramp_time_over;
249
250 /*
251 * Time since last latency_window was started
252 */
253 struct timeval latency_ts;
254 unsigned int latency_qd;
255 unsigned int latency_qd_high;
256 unsigned int latency_qd_low;
257 unsigned int latency_failed;
258 uint64_t latency_ios;
259 int latency_end_run;
260
261 /*
262 * read/write mixed workload state
263 */
264 struct frand_state rwmix_state;
265 unsigned long rwmix_issues;
266 enum fio_ddir rwmix_ddir;
267 unsigned int ddir_seq_nr;
268
269 /*
270 * rand/seq mixed workload state
271 */
272 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
273
274 /*
275 * IO history logs for verification. We use a tree for sorting,
276 * if we are overwriting. Otherwise just use a fifo.
277 */
278 struct rb_root io_hist_tree;
279 struct flist_head io_hist_list;
280 unsigned long io_hist_len;
281
282 /*
283 * For IO replaying
284 */
285 struct flist_head io_log_list;
286
287 /*
288 * For tracking/handling discards
289 */
290 struct flist_head trim_list;
291 unsigned long trim_entries;
292
293 struct flist_head next_rand_list;
294
295 /*
296 * for fileservice, how often to switch to a new file
297 */
298 unsigned int file_service_nr;
299 unsigned int file_service_left;
300 struct fio_file *file_service_file;
301
302 unsigned int sync_file_range_nr;
303
304 /*
305 * For generating file sizes
306 */
307 struct frand_state file_size_state;
308
309 /*
310 * Error counts
311 */
312 unsigned int total_err_count;
313 int first_error;
314
315 struct fio_flow *flow;
316
317 /*
318 * Can be overloaded by profiles
319 */
320 struct prof_io_ops prof_io_ops;
321 void *prof_data;
322
323 void *pinned_mem;
324};
325
326/*
327 * when should interactive ETA output be generated
328 */
329enum {
330 FIO_ETA_AUTO,
331 FIO_ETA_ALWAYS,
332 FIO_ETA_NEVER,
333};
334
335#define __td_verror(td, err, msg, func) \
336 do { \
337 unsigned int ____e = (err); \
338 if ((td)->error) \
339 break; \
340 (td)->error = ____e; \
341 if (!(td)->first_error) \
342 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
343 } while (0)
344
345
346#define td_clear_error(td) \
347 (td)->error = 0;
348#define td_verror(td, err, func) \
349 __td_verror((td), (err), strerror((err)), (func))
350#define td_vmsg(td, err, msg, func) \
351 __td_verror((td), (err), (msg), (func))
352
353#define __fio_stringify_1(x) #x
354#define __fio_stringify(x) __fio_stringify_1(x)
355
356extern int exitall_on_terminate;
357extern unsigned int thread_number;
358extern unsigned int stat_number;
359extern int shm_id;
360extern int groupid;
361extern int output_format;
362extern int append_terse_output;
363extern int temp_stall_ts;
364extern uintptr_t page_mask, page_size;
365extern int read_only;
366extern int eta_print;
367extern int eta_new_line;
368extern unsigned long done_secs;
369extern char *job_section;
370extern int fio_gtod_offload;
371extern int fio_gtod_cpu;
372extern enum fio_cs fio_clock_source;
373extern int fio_clock_source_set;
374extern int warnings_fatal;
375extern int terse_version;
376extern int is_backend;
377extern int nr_clients;
378extern int log_syslog;
379extern int status_interval;
380extern const char fio_version_string[];
381extern int helper_do_stat;
382extern pthread_cond_t helper_cond;
383
384extern struct thread_data *threads;
385
386static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
387{
388 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
389}
390
391#define REAL_MAX_JOBS 2048
392
393static inline int should_fsync(struct thread_data *td)
394{
395 if (td->last_was_sync)
396 return 0;
397 if (td_write(td) || td_rw(td) || td->o.override_sync)
398 return 1;
399
400 return 0;
401}
402
403/*
404 * Init/option functions
405 */
406extern int __must_check fio_init_options(void);
407extern int __must_check parse_options(int, char **);
408extern int parse_jobs_ini(char *, int, int, int);
409extern int parse_cmd_line(int, char **, int);
410extern int fio_backend(void);
411extern void reset_fio_state(void);
412extern void clear_io_state(struct thread_data *);
413extern int fio_options_parse(struct thread_data *, char **, int, int);
414extern void fio_keywords_init(void);
415extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
416extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
417extern void fio_fill_default_options(struct thread_data *);
418extern int fio_show_option_help(const char *);
419extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
420extern void fio_options_dup_and_init(struct option *);
421extern void fio_options_mem_dupe(struct thread_data *);
422extern void options_mem_dupe(void *data, struct fio_option *options);
423extern void td_fill_rand_seeds(struct thread_data *);
424extern void add_job_opts(const char **, int);
425extern char *num2str(unsigned long, int, int, int, int);
426extern int ioengine_load(struct thread_data *);
427extern int parse_dryrun(void);
428extern int fio_running_or_pending_io_threads(void);
429extern int fio_set_fd_nonblocking(int, const char *);
430
431extern uintptr_t page_mask;
432extern uintptr_t page_size;
433extern int initialize_fio(char *envp[]);
434
435#define FIO_GETOPT_JOB 0x89000000
436#define FIO_GETOPT_IOENGINE 0x98000000
437#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
438
439/*
440 * ETA/status stuff
441 */
442extern void print_thread_status(void);
443extern void print_status_init(int);
444extern char *fio_uint_to_kmg(unsigned int val);
445
446/*
447 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
448 * will never back again. It may cycle between running/verififying/fsyncing.
449 * Once the thread reaches TD_EXITED, it is just waiting for the core to
450 * reap it.
451 */
452enum {
453 TD_NOT_CREATED = 0,
454 TD_CREATED,
455 TD_INITIALIZED,
456 TD_RAMP,
457 TD_SETTING_UP,
458 TD_RUNNING,
459 TD_PRE_READING,
460 TD_VERIFYING,
461 TD_FSYNCING,
462 TD_FINISHING,
463 TD_EXITED,
464 TD_REAPED,
465};
466
467extern void td_set_runstate(struct thread_data *, int);
468extern int td_bump_runstate(struct thread_data *, int);
469extern void td_restore_runstate(struct thread_data *, int);
470
471/*
472 * Allow 60 seconds for a job to quit on its own, otherwise reap with
473 * a vengeance.
474 */
475#define FIO_REAP_TIMEOUT 60
476
477#define TERMINATE_ALL (-1)
478extern void fio_terminate_threads(int);
479extern void fio_mark_td_terminate(struct thread_data *);
480
481/*
482 * Memory helpers
483 */
484extern int __must_check fio_pin_memory(struct thread_data *);
485extern void fio_unpin_memory(struct thread_data *);
486extern int __must_check allocate_io_mem(struct thread_data *);
487extern void free_io_mem(struct thread_data *);
488extern void free_threads_shm(void);
489
490/*
491 * Reset stats after ramp time completes
492 */
493extern void reset_all_stats(struct thread_data *);
494
495/*
496 * blktrace support
497 */
498#ifdef FIO_HAVE_BLKTRACE
499extern int is_blktrace(const char *, int *);
500extern int load_blktrace(struct thread_data *, const char *, int);
501#endif
502
503/*
504 * Latency target helpers
505 */
506extern void lat_target_check(struct thread_data *);
507extern void lat_target_init(struct thread_data *);
508extern void lat_target_reset(struct thread_data *);
509
510#define for_each_td(td, i) \
511 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
512#define for_each_file(td, f, i) \
513 if ((td)->files_index) \
514 for ((i) = 0, (f) = (td)->files[0]; \
515 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
516 (i)++)
517
518#define fio_assert(td, cond) do { \
519 if (!(cond)) { \
520 int *__foo = NULL; \
521 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
522 td_set_runstate((td), TD_EXITED); \
523 (td)->error = EFAULT; \
524 *__foo = 0; \
525 } \
526} while (0)
527
528static inline int fio_fill_issue_time(struct thread_data *td)
529{
530 if (td->o.read_iolog_file ||
531 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
532 return 1;
533
534 return 0;
535}
536
537static inline int __should_check_rate(struct thread_data *td,
538 enum fio_ddir ddir)
539{
540 struct thread_options *o = &td->o;
541
542 /*
543 * If some rate setting was given, we need to check it
544 */
545 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
546 o->rate_iops_min[ddir])
547 return 1;
548
549 return 0;
550}
551
552static inline int should_check_rate(struct thread_data *td,
553 uint64_t *bytes_done)
554{
555 int ret = 0;
556
557 if (bytes_done[DDIR_READ])
558 ret |= __should_check_rate(td, DDIR_READ);
559 if (bytes_done[DDIR_WRITE])
560 ret |= __should_check_rate(td, DDIR_WRITE);
561 if (bytes_done[DDIR_TRIM])
562 ret |= __should_check_rate(td, DDIR_TRIM);
563
564 return ret;
565}
566
567static inline unsigned int td_max_bs(struct thread_data *td)
568{
569 unsigned int max_bs;
570
571 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
572 return max(td->o.max_bs[DDIR_TRIM], max_bs);
573}
574
575static inline unsigned int td_min_bs(struct thread_data *td)
576{
577 unsigned int min_bs;
578
579 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
580 return min(td->o.min_bs[DDIR_TRIM], min_bs);
581}
582
583static inline int is_power_of_2(unsigned long val)
584{
585 return (val != 0 && ((val & (val - 1)) == 0));
586}
587
588/*
589 * We currently only need to do locking if we have verifier threads
590 * accessing our internal structures too
591 */
592static inline void td_io_u_lock(struct thread_data *td)
593{
594 if (td->o.verify_async)
595 pthread_mutex_lock(&td->io_u_lock);
596}
597
598static inline void td_io_u_unlock(struct thread_data *td)
599{
600 if (td->o.verify_async)
601 pthread_mutex_unlock(&td->io_u_lock);
602}
603
604static inline void td_io_u_free_notify(struct thread_data *td)
605{
606 if (td->o.verify_async)
607 pthread_cond_signal(&td->free_cond);
608}
609
610extern const char *fio_get_arch_string(int);
611extern const char *fio_get_os_string(int);
612
613#ifdef FIO_INTERNAL
614#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
615#endif
616
617enum {
618 FIO_OUTPUT_TERSE = 0,
619 FIO_OUTPUT_JSON,
620 FIO_OUTPUT_NORMAL,
621};
622
623enum {
624 FIO_RAND_DIST_RANDOM = 0,
625 FIO_RAND_DIST_ZIPF,
626 FIO_RAND_DIST_PARETO,
627};
628
629enum {
630 FIO_RAND_GEN_TAUSWORTHE = 0,
631 FIO_RAND_GEN_LFSR,
632};
633
634enum {
635 FIO_CPUS_SHARED = 0,
636 FIO_CPUS_SPLIT,
637};
638
639#endif
fio - Flexible IO Tester