Fixup bw/iops logging for short runs
[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#include "workqueue.h"
44
45#ifdef CONFIG_SOLARISAIO
46#include <sys/asynch.h>
47#endif
48
49#ifdef CONFIG_LIBNUMA
50#include <linux/mempolicy.h>
51#include <numa.h>
52
53/*
54 * "local" is pseudo-policy
55 */
56#define MPOL_LOCAL MPOL_MAX
57#endif
58
59/*
60 * offset generator types
61 */
62enum {
63 RW_SEQ_SEQ = 0,
64 RW_SEQ_IDENT,
65};
66
67enum {
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,
76 TD_F_NOIO = 1U << 8,
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};
83
84enum {
85 FIO_RAND_BS_OFF = 0,
86 FIO_RAND_VER_OFF,
87 FIO_RAND_MIX_OFF,
88 FIO_RAND_FILE_OFF,
89 FIO_RAND_BLOCK_OFF,
90 FIO_RAND_FILE_SIZE_OFF,
91 FIO_RAND_TRIM_OFF,
92 FIO_RAND_BUF_OFF,
93 FIO_RAND_SEQ_RAND_READ_OFF,
94 FIO_RAND_SEQ_RAND_WRITE_OFF,
95 FIO_RAND_SEQ_RAND_TRIM_OFF,
96 FIO_RAND_START_DELAY,
97 FIO_DEDUPE_OFF,
98 FIO_RAND_NR_OFFS,
99};
100
101enum {
102 IO_MODE_INLINE = 0,
103 IO_MODE_OFFLOAD,
104};
105
106/*
107 * This describes a single thread/process executing a fio job.
108 */
109struct thread_data {
110 struct thread_options o;
111 unsigned long flags;
112 void *eo;
113 char verror[FIO_VERROR_SIZE];
114 pthread_t thread;
115 unsigned int thread_number;
116 unsigned int subjob_number;
117 unsigned int groupid;
118 struct thread_stat ts;
119
120 int client_type;
121
122 struct io_log *slat_log;
123 struct io_log *clat_log;
124 struct io_log *lat_log;
125 struct io_log *bw_log;
126 struct io_log *iops_log;
127
128 struct tp_data *tp_data;
129
130 struct thread_data *parent;
131
132 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
133 struct timeval bw_sample_time;
134
135 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
136 struct timeval iops_sample_time;
137
138 /*
139 * Tracks the last iodepth number of completed writes, if data
140 * verification is enabled
141 */
142 uint64_t *last_write_comp;
143 unsigned int last_write_idx;
144
145 volatile int update_rusage;
146 struct fio_mutex *rusage_sem;
147 struct rusage ru_start;
148 struct rusage ru_end;
149
150 struct fio_file **files;
151 unsigned char *file_locks;
152 unsigned int files_size;
153 unsigned int files_index;
154 unsigned int nr_open_files;
155 unsigned int nr_done_files;
156 unsigned int nr_normal_files;
157 union {
158 unsigned int next_file;
159 struct frand_state next_file_state;
160 };
161 int error;
162 int sig;
163 int done;
164 int stop_io;
165 pid_t pid;
166 char *orig_buffer;
167 size_t orig_buffer_size;
168 volatile int terminate;
169 volatile int runstate;
170 unsigned int last_was_sync;
171 enum fio_ddir last_ddir;
172
173 int mmapfd;
174
175 void *iolog_buf;
176 FILE *iolog_f;
177
178 char *sysfs_root;
179
180 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
181
182 struct frand_state bsrange_state;
183 struct frand_state verify_state;
184 struct frand_state trim_state;
185 struct frand_state delay_state;
186
187 struct frand_state buf_state;
188 struct frand_state buf_state_prev;
189 struct frand_state dedupe_state;
190
191 unsigned int verify_batch;
192 unsigned int trim_batch;
193
194 struct thread_io_list *vstate;
195
196 int shm_id;
197
198 /*
199 * IO engine hooks, contains everything needed to submit an io_u
200 * to any of the available IO engines.
201 */
202 struct ioengine_ops *io_ops;
203
204 /*
205 * Queue depth of io_u's that fio MIGHT do
206 */
207 unsigned int cur_depth;
208
209 /*
210 * io_u's about to be committed
211 */
212 unsigned int io_u_queued;
213
214 /*
215 * io_u's submitted but not completed yet
216 */
217 unsigned int io_u_in_flight;
218
219 /*
220 * List of free and busy io_u's
221 */
222 struct io_u_ring io_u_requeues;
223 struct io_u_queue io_u_freelist;
224 struct io_u_queue io_u_all;
225 pthread_mutex_t io_u_lock;
226 pthread_cond_t free_cond;
227
228 /*
229 * async verify offload
230 */
231 struct flist_head verify_list;
232 pthread_t *verify_threads;
233 unsigned int nr_verify_threads;
234 pthread_cond_t verify_cond;
235 int verify_thread_exit;
236
237 /*
238 * Rate state
239 */
240 uint64_t rate_bps[DDIR_RWDIR_CNT];
241 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
242 unsigned long rate_bytes[DDIR_RWDIR_CNT];
243 unsigned long rate_blocks[DDIR_RWDIR_CNT];
244 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
245 struct timeval lastrate[DDIR_RWDIR_CNT];
246
247 /*
248 * Enforced rate submission/completion workqueue
249 */
250 struct workqueue io_wq;
251
252 uint64_t total_io_size;
253 uint64_t fill_device_size;
254
255 /*
256 * Issue side
257 */
258 uint64_t io_issues[DDIR_RWDIR_CNT];
259 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
260 uint64_t loops;
261
262 /*
263 * Completions
264 */
265 uint64_t io_blocks[DDIR_RWDIR_CNT];
266 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
267 uint64_t io_bytes[DDIR_RWDIR_CNT];
268 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
269 uint64_t io_skip_bytes;
270 uint64_t zone_bytes;
271 struct fio_mutex *mutex;
272 uint64_t bytes_done[DDIR_RWDIR_CNT];
273
274 /*
275 * State for random io, a bitmap of blocks done vs not done
276 */
277 struct frand_state random_state;
278
279 struct timeval start; /* start of this loop */
280 struct timeval epoch; /* time job was started */
281 struct timeval last_issue;
282 long time_offset;
283 struct timeval tv_cache;
284 struct timeval terminate_time;
285 unsigned int tv_cache_nr;
286 unsigned int tv_cache_mask;
287 unsigned int ramp_time_over;
288
289 /*
290 * Time since last latency_window was started
291 */
292 struct timeval latency_ts;
293 unsigned int latency_qd;
294 unsigned int latency_qd_high;
295 unsigned int latency_qd_low;
296 unsigned int latency_failed;
297 uint64_t latency_ios;
298 int latency_end_run;
299
300 /*
301 * read/write mixed workload state
302 */
303 struct frand_state rwmix_state;
304 unsigned long rwmix_issues;
305 enum fio_ddir rwmix_ddir;
306 unsigned int ddir_seq_nr;
307
308 /*
309 * rand/seq mixed workload state
310 */
311 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
312
313 /*
314 * IO history logs for verification. We use a tree for sorting,
315 * if we are overwriting. Otherwise just use a fifo.
316 */
317 struct rb_root io_hist_tree;
318 struct flist_head io_hist_list;
319 unsigned long io_hist_len;
320
321 /*
322 * For IO replaying
323 */
324 struct flist_head io_log_list;
325
326 /*
327 * For tracking/handling discards
328 */
329 struct flist_head trim_list;
330 unsigned long trim_entries;
331
332 struct flist_head next_rand_list;
333
334 /*
335 * for fileservice, how often to switch to a new file
336 */
337 unsigned int file_service_nr;
338 unsigned int file_service_left;
339 struct fio_file *file_service_file;
340
341 unsigned int sync_file_range_nr;
342
343 /*
344 * For generating file sizes
345 */
346 struct frand_state file_size_state;
347
348 /*
349 * Error counts
350 */
351 unsigned int total_err_count;
352 int first_error;
353
354 struct fio_flow *flow;
355
356 /*
357 * Can be overloaded by profiles
358 */
359 struct prof_io_ops prof_io_ops;
360 void *prof_data;
361
362 void *pinned_mem;
363};
364
365/*
366 * when should interactive ETA output be generated
367 */
368enum {
369 FIO_ETA_AUTO,
370 FIO_ETA_ALWAYS,
371 FIO_ETA_NEVER,
372};
373
374#define __td_verror(td, err, msg, func) \
375 do { \
376 unsigned int ____e = (err); \
377 if ((td)->error) \
378 break; \
379 (td)->error = ____e; \
380 if (!(td)->first_error) \
381 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
382 } while (0)
383
384
385#define td_clear_error(td) do { \
386 (td)->error = 0; \
387 if ((td)->parent) \
388 (td)->parent->error = 0; \
389} while (0)
390
391#define td_verror(td, err, func) do { \
392 __td_verror((td), (err), strerror((err)), (func)); \
393 if ((td)->parent) \
394 __td_verror((td)->parent, (err), strerror((err)), (func)); \
395} while (0)
396
397#define td_vmsg(td, err, msg, func) do { \
398 __td_verror((td), (err), (msg), (func)); \
399 if ((td)->parent) \
400 __td_verror((td)->parent, (err), (msg), (func)); \
401} while (0)
402
403#define __fio_stringify_1(x) #x
404#define __fio_stringify(x) __fio_stringify_1(x)
405
406extern int exitall_on_terminate;
407extern unsigned int thread_number;
408extern unsigned int stat_number;
409extern int shm_id;
410extern int groupid;
411extern int output_format;
412extern int append_terse_output;
413extern int temp_stall_ts;
414extern uintptr_t page_mask, page_size;
415extern int read_only;
416extern int eta_print;
417extern int eta_new_line;
418extern unsigned long done_secs;
419extern char *job_section;
420extern int fio_gtod_offload;
421extern int fio_gtod_cpu;
422extern enum fio_cs fio_clock_source;
423extern int fio_clock_source_set;
424extern int warnings_fatal;
425extern int terse_version;
426extern int is_backend;
427extern int nr_clients;
428extern int log_syslog;
429extern int status_interval;
430extern const char fio_version_string[];
431extern int helper_do_stat;
432extern pthread_cond_t helper_cond;
433extern char *trigger_file;
434extern char *trigger_cmd;
435extern char *trigger_remote_cmd;
436extern long long trigger_timeout;
437extern char *aux_path;
438
439extern struct thread_data *threads;
440
441static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
442{
443 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
444}
445
446#define REAL_MAX_JOBS 2048
447
448static inline int should_fsync(struct thread_data *td)
449{
450 if (td->last_was_sync)
451 return 0;
452 if (td_write(td) || td_rw(td) || td->o.override_sync)
453 return 1;
454
455 return 0;
456}
457
458/*
459 * Init/option functions
460 */
461extern int __must_check fio_init_options(void);
462extern int __must_check parse_options(int, char **);
463extern int parse_jobs_ini(char *, int, int, int);
464extern int parse_cmd_line(int, char **, int);
465extern int fio_backend(void);
466extern void reset_fio_state(void);
467extern void clear_io_state(struct thread_data *, int);
468extern int fio_options_parse(struct thread_data *, char **, int, int);
469extern void fio_keywords_init(void);
470extern void fio_keywords_exit(void);
471extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
472extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
473extern void fio_fill_default_options(struct thread_data *);
474extern int fio_show_option_help(const char *);
475extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
476extern void fio_options_dup_and_init(struct option *);
477extern void fio_options_mem_dupe(struct thread_data *);
478extern void options_mem_dupe(void *data, struct fio_option *options);
479extern void td_fill_rand_seeds(struct thread_data *);
480extern void add_job_opts(const char **, int);
481extern char *num2str(uint64_t, int, int, int, int);
482extern int ioengine_load(struct thread_data *);
483extern int parse_dryrun(void);
484extern int fio_running_or_pending_io_threads(void);
485extern int fio_set_fd_nonblocking(int, const char *);
486extern void sig_show_status(int sig);
487
488extern uintptr_t page_mask;
489extern uintptr_t page_size;
490extern int initialize_fio(char *envp[]);
491extern void deinitialize_fio(void);
492
493#define FIO_GETOPT_JOB 0x89000000
494#define FIO_GETOPT_IOENGINE 0x98000000
495#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
496
497/*
498 * ETA/status stuff
499 */
500extern void print_thread_status(void);
501extern void print_status_init(int);
502extern char *fio_uint_to_kmg(unsigned int val);
503
504/*
505 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
506 * will never back again. It may cycle between running/verififying/fsyncing.
507 * Once the thread reaches TD_EXITED, it is just waiting for the core to
508 * reap it.
509 */
510enum {
511 TD_NOT_CREATED = 0,
512 TD_CREATED,
513 TD_INITIALIZED,
514 TD_RAMP,
515 TD_SETTING_UP,
516 TD_RUNNING,
517 TD_PRE_READING,
518 TD_VERIFYING,
519 TD_FSYNCING,
520 TD_FINISHING,
521 TD_EXITED,
522 TD_REAPED,
523 TD_LAST,
524};
525
526extern void td_set_runstate(struct thread_data *, int);
527extern int td_bump_runstate(struct thread_data *, int);
528extern void td_restore_runstate(struct thread_data *, int);
529
530/*
531 * Allow 60 seconds for a job to quit on its own, otherwise reap with
532 * a vengeance.
533 */
534#define FIO_REAP_TIMEOUT 60
535
536#define TERMINATE_ALL (-1)
537extern void fio_terminate_threads(int);
538extern void fio_mark_td_terminate(struct thread_data *);
539
540/*
541 * Memory helpers
542 */
543extern int __must_check fio_pin_memory(struct thread_data *);
544extern void fio_unpin_memory(struct thread_data *);
545extern int __must_check allocate_io_mem(struct thread_data *);
546extern void free_io_mem(struct thread_data *);
547extern void free_threads_shm(void);
548
549/*
550 * Reset stats after ramp time completes
551 */
552extern void reset_all_stats(struct thread_data *);
553
554/*
555 * blktrace support
556 */
557#ifdef FIO_HAVE_BLKTRACE
558extern int is_blktrace(const char *, int *);
559extern int load_blktrace(struct thread_data *, const char *, int);
560#endif
561
562/*
563 * Latency target helpers
564 */
565extern void lat_target_check(struct thread_data *);
566extern void lat_target_init(struct thread_data *);
567extern void lat_target_reset(struct thread_data *);
568
569#define for_each_td(td, i) \
570 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
571#define for_each_file(td, f, i) \
572 if ((td)->files_index) \
573 for ((i) = 0, (f) = (td)->files[0]; \
574 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
575 (i)++)
576
577#define fio_assert(td, cond) do { \
578 if (!(cond)) { \
579 int *__foo = NULL; \
580 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
581 td_set_runstate((td), TD_EXITED); \
582 (td)->error = EFAULT; \
583 *__foo = 0; \
584 } \
585} while (0)
586
587static inline int fio_fill_issue_time(struct thread_data *td)
588{
589 if (td->o.read_iolog_file ||
590 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
591 return 1;
592
593 return 0;
594}
595
596static inline int __should_check_rate(struct thread_data *td,
597 enum fio_ddir ddir)
598{
599 struct thread_options *o = &td->o;
600
601 /*
602 * If some rate setting was given, we need to check it
603 */
604 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
605 o->rate_iops_min[ddir])
606 return 1;
607
608 return 0;
609}
610
611static inline int should_check_rate(struct thread_data *td)
612{
613 int ret = 0;
614
615 if (td->bytes_done[DDIR_READ])
616 ret |= __should_check_rate(td, DDIR_READ);
617 if (td->bytes_done[DDIR_WRITE])
618 ret |= __should_check_rate(td, DDIR_WRITE);
619 if (td->bytes_done[DDIR_TRIM])
620 ret |= __should_check_rate(td, DDIR_TRIM);
621
622 return ret;
623}
624
625static inline unsigned int td_max_bs(struct thread_data *td)
626{
627 unsigned int max_bs;
628
629 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
630 return max(td->o.max_bs[DDIR_TRIM], max_bs);
631}
632
633static inline unsigned int td_min_bs(struct thread_data *td)
634{
635 unsigned int min_bs;
636
637 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
638 return min(td->o.min_bs[DDIR_TRIM], min_bs);
639}
640
641static inline int td_async_processing(struct thread_data *td)
642{
643 return (td->flags & TD_F_NEED_LOCK) != 0;
644}
645
646/*
647 * We currently only need to do locking if we have verifier threads
648 * accessing our internal structures too
649 */
650static inline void td_io_u_lock(struct thread_data *td)
651{
652 if (td_async_processing(td))
653 pthread_mutex_lock(&td->io_u_lock);
654}
655
656static inline void td_io_u_unlock(struct thread_data *td)
657{
658 if (td_async_processing(td))
659 pthread_mutex_unlock(&td->io_u_lock);
660}
661
662static inline void td_io_u_free_notify(struct thread_data *td)
663{
664 if (td_async_processing(td))
665 pthread_cond_signal(&td->free_cond);
666}
667
668extern const char *fio_get_arch_string(int);
669extern const char *fio_get_os_string(int);
670
671#ifdef FIO_INTERNAL
672#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
673#define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
674#endif
675
676enum {
677 __FIO_OUTPUT_TERSE = 0,
678 __FIO_OUTPUT_JSON = 1,
679 __FIO_OUTPUT_NORMAL = 2,
680 FIO_OUTPUT_NR = 3,
681
682 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
683 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
684 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
685};
686
687enum {
688 FIO_RAND_DIST_RANDOM = 0,
689 FIO_RAND_DIST_ZIPF,
690 FIO_RAND_DIST_PARETO,
691 FIO_RAND_DIST_GAUSS,
692};
693
694#define FIO_DEF_ZIPF 1.1
695#define FIO_DEF_PARETO 0.2
696
697enum {
698 FIO_RAND_GEN_TAUSWORTHE = 0,
699 FIO_RAND_GEN_LFSR,
700 FIO_RAND_GEN_TAUSWORTHE64,
701};
702
703enum {
704 FIO_CPUS_SHARED = 0,
705 FIO_CPUS_SPLIT,
706};
707
708extern void exec_trigger(const char *);
709extern void check_trigger_file(void);
710
711#endif