btrace2fio: expand rate options, bug fixes
[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_RAND_NR_OFFS,
93};
94
95/*
96 * This describes a single thread/process executing a fio job.
97 */
98struct thread_data {
99 struct thread_options o;
100 unsigned long flags;
101 void *eo;
102 char verror[FIO_VERROR_SIZE];
103 pthread_t thread;
104 unsigned int thread_number;
105 unsigned int subjob_number;
106 unsigned int groupid;
107 struct thread_stat ts;
108
109 int client_type;
110
111 struct io_log *slat_log;
112 struct io_log *clat_log;
113 struct io_log *lat_log;
114 struct io_log *bw_log;
115 struct io_log *iops_log;
116
117 struct tp_data *tp_data;
118
119 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
120 struct timeval bw_sample_time;
121
122 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
123 struct timeval iops_sample_time;
124
125 volatile int update_rusage;
126 struct fio_mutex *rusage_sem;
127 struct rusage ru_start;
128 struct rusage ru_end;
129
130 struct fio_file **files;
131 unsigned char *file_locks;
132 unsigned int files_size;
133 unsigned int files_index;
134 unsigned int nr_open_files;
135 unsigned int nr_done_files;
136 unsigned int nr_normal_files;
137 union {
138 unsigned int next_file;
139 os_random_state_t next_file_state;
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 union {
163 os_random_state_t bsrange_state;
164 struct frand_state __bsrange_state;
165 };
166 union {
167 os_random_state_t verify_state;
168 struct frand_state __verify_state;
169 };
170 union {
171 os_random_state_t trim_state;
172 struct frand_state __trim_state;
173 };
174 union {
175 os_random_state_t delay_state;
176 struct frand_state __delay_state;
177 };
178
179 struct frand_state buf_state;
180
181 unsigned int verify_batch;
182 unsigned int trim_batch;
183
184 int shm_id;
185
186 /*
187 * IO engine hooks, contains everything needed to submit an io_u
188 * to any of the available IO engines.
189 */
190 struct ioengine_ops *io_ops;
191
192 /*
193 * Queue depth of io_u's that fio MIGHT do
194 */
195 unsigned int cur_depth;
196
197 /*
198 * io_u's about to be committed
199 */
200 unsigned int io_u_queued;
201
202 /*
203 * io_u's submitted but not completed yet
204 */
205 unsigned int io_u_in_flight;
206
207 /*
208 * List of free and busy io_u's
209 */
210 struct io_u_ring io_u_requeues;
211 struct io_u_queue io_u_freelist;
212 struct io_u_queue io_u_all;
213 pthread_mutex_t io_u_lock;
214 pthread_cond_t free_cond;
215
216 /*
217 * async verify offload
218 */
219 struct flist_head verify_list;
220 pthread_t *verify_threads;
221 unsigned int nr_verify_threads;
222 pthread_cond_t verify_cond;
223 int verify_thread_exit;
224
225 /*
226 * Rate state
227 */
228 uint64_t rate_bps[DDIR_RWDIR_CNT];
229 long rate_pending_usleep[DDIR_RWDIR_CNT];
230 unsigned long rate_bytes[DDIR_RWDIR_CNT];
231 unsigned long rate_blocks[DDIR_RWDIR_CNT];
232 struct timeval lastrate[DDIR_RWDIR_CNT];
233
234 uint64_t total_io_size;
235 uint64_t fill_device_size;
236
237 unsigned long io_issues[DDIR_RWDIR_CNT];
238 uint64_t io_blocks[DDIR_RWDIR_CNT];
239 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
240 uint64_t io_bytes[DDIR_RWDIR_CNT];
241 uint64_t io_skip_bytes;
242 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
243 uint64_t zone_bytes;
244 struct fio_mutex *mutex;
245
246 /*
247 * State for random io, a bitmap of blocks done vs not done
248 */
249 union {
250 os_random_state_t random_state;
251 struct frand_state __random_state;
252 };
253
254 struct timeval start; /* start of this loop */
255 struct timeval epoch; /* time job was started */
256 struct timeval last_issue;
257 struct timeval tv_cache;
258 struct timeval terminate_time;
259 unsigned int tv_cache_nr;
260 unsigned int tv_cache_mask;
261 unsigned int ramp_time_over;
262
263 /*
264 * Time since last latency_window was started
265 */
266 struct timeval latency_ts;
267 unsigned int latency_qd;
268 unsigned int latency_qd_high;
269 unsigned int latency_qd_low;
270 unsigned int latency_failed;
271 uint64_t latency_ios;
272 int latency_end_run;
273
274 /*
275 * read/write mixed workload state
276 */
277 union {
278 os_random_state_t rwmix_state;
279 struct frand_state __rwmix_state;
280 };
281 unsigned long rwmix_issues;
282 enum fio_ddir rwmix_ddir;
283 unsigned int ddir_seq_nr;
284
285 /*
286 * rand/seq mixed workload state
287 */
288 union {
289 os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
290 struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
291 };
292
293 /*
294 * IO history logs for verification. We use a tree for sorting,
295 * if we are overwriting. Otherwise just use a fifo.
296 */
297 struct rb_root io_hist_tree;
298 struct flist_head io_hist_list;
299 unsigned long io_hist_len;
300
301 /*
302 * For IO replaying
303 */
304 struct flist_head io_log_list;
305
306 /*
307 * For tracking/handling discards
308 */
309 struct flist_head trim_list;
310 unsigned long trim_entries;
311
312 struct flist_head next_rand_list;
313
314 /*
315 * for fileservice, how often to switch to a new file
316 */
317 unsigned int file_service_nr;
318 unsigned int file_service_left;
319 struct fio_file *file_service_file;
320
321 unsigned int sync_file_range_nr;
322
323 /*
324 * For generating file sizes
325 */
326 union {
327 os_random_state_t file_size_state;
328 struct frand_state __file_size_state;
329 };
330
331 /*
332 * Error counts
333 */
334 unsigned int total_err_count;
335 int first_error;
336
337 struct fio_flow *flow;
338
339 /*
340 * Can be overloaded by profiles
341 */
342 struct prof_io_ops prof_io_ops;
343 void *prof_data;
344
345 void *pinned_mem;
346};
347
348/*
349 * when should interactive ETA output be generated
350 */
351enum {
352 FIO_ETA_AUTO,
353 FIO_ETA_ALWAYS,
354 FIO_ETA_NEVER,
355};
356
357#define __td_verror(td, err, msg, func) \
358 do { \
359 unsigned int ____e = (err); \
360 if ((td)->error) \
361 break; \
362 (td)->error = ____e; \
363 if (!(td)->first_error) \
364 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
365 } while (0)
366
367
368#define td_clear_error(td) \
369 (td)->error = 0;
370#define td_verror(td, err, func) \
371 __td_verror((td), (err), strerror((err)), (func))
372#define td_vmsg(td, err, msg, func) \
373 __td_verror((td), (err), (msg), (func))
374
375#define __fio_stringify_1(x) #x
376#define __fio_stringify(x) __fio_stringify_1(x)
377
378extern int exitall_on_terminate;
379extern unsigned int thread_number;
380extern unsigned int stat_number;
381extern int shm_id;
382extern int groupid;
383extern int output_format;
384extern int append_terse_output;
385extern int temp_stall_ts;
386extern uintptr_t page_mask, page_size;
387extern int read_only;
388extern int eta_print;
389extern int eta_new_line;
390extern unsigned long done_secs;
391extern char *job_section;
392extern int fio_gtod_offload;
393extern int fio_gtod_cpu;
394extern enum fio_cs fio_clock_source;
395extern int fio_clock_source_set;
396extern int warnings_fatal;
397extern int terse_version;
398extern int is_backend;
399extern int nr_clients;
400extern int log_syslog;
401extern int status_interval;
402extern const char fio_version_string[];
403
404extern struct thread_data *threads;
405
406static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
407{
408 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
409}
410
411#define REAL_MAX_JOBS 2048
412
413static inline int should_fsync(struct thread_data *td)
414{
415 if (td->last_was_sync)
416 return 0;
417 if (td_write(td) || td_rw(td) || td->o.override_sync)
418 return 1;
419
420 return 0;
421}
422
423/*
424 * Init/option functions
425 */
426extern int __must_check fio_init_options(void);
427extern int __must_check parse_options(int, char **);
428extern int parse_jobs_ini(char *, int, int, int);
429extern int parse_cmd_line(int, char **, int);
430extern int fio_backend(void);
431extern void reset_fio_state(void);
432extern void clear_io_state(struct thread_data *);
433extern int fio_options_parse(struct thread_data *, char **, int, int);
434extern void fio_keywords_init(void);
435extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
436extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
437extern void fio_fill_default_options(struct thread_data *);
438extern int fio_show_option_help(const char *);
439extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
440extern void fio_options_dup_and_init(struct option *);
441extern void fio_options_mem_dupe(struct thread_data *);
442extern void options_mem_dupe(void *data, struct fio_option *options);
443extern void td_fill_rand_seeds(struct thread_data *);
444extern void add_job_opts(const char **, int);
445extern char *num2str(unsigned long, int, int, int, int);
446extern int ioengine_load(struct thread_data *);
447extern int parse_dryrun(void);
448extern int fio_running_or_pending_io_threads(void);
449extern int fio_set_fd_nonblocking(int, const char *);
450
451extern uintptr_t page_mask;
452extern uintptr_t page_size;
453extern int initialize_fio(char *envp[]);
454
455#define FIO_GETOPT_JOB 0x89000000
456#define FIO_GETOPT_IOENGINE 0x98000000
457#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
458
459/*
460 * ETA/status stuff
461 */
462extern void print_thread_status(void);
463extern void print_status_init(int);
464extern char *fio_uint_to_kmg(unsigned int val);
465
466/*
467 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
468 * will never back again. It may cycle between running/verififying/fsyncing.
469 * Once the thread reaches TD_EXITED, it is just waiting for the core to
470 * reap it.
471 */
472enum {
473 TD_NOT_CREATED = 0,
474 TD_CREATED,
475 TD_INITIALIZED,
476 TD_RAMP,
477 TD_SETTING_UP,
478 TD_RUNNING,
479 TD_PRE_READING,
480 TD_VERIFYING,
481 TD_FSYNCING,
482 TD_FINISHING,
483 TD_EXITED,
484 TD_REAPED,
485};
486
487extern void td_set_runstate(struct thread_data *, int);
488extern int td_bump_runstate(struct thread_data *, int);
489extern void td_restore_runstate(struct thread_data *, int);
490
491/*
492 * Allow 60 seconds for a job to quit on its own, otherwise reap with
493 * a vengeance.
494 */
495#define FIO_REAP_TIMEOUT 60
496
497#define TERMINATE_ALL (-1)
498extern void fio_terminate_threads(int);
499extern void fio_mark_td_terminate(struct thread_data *);
500
501/*
502 * Memory helpers
503 */
504extern int __must_check fio_pin_memory(struct thread_data *);
505extern void fio_unpin_memory(struct thread_data *);
506extern int __must_check allocate_io_mem(struct thread_data *);
507extern void free_io_mem(struct thread_data *);
508extern void free_threads_shm(void);
509
510/*
511 * Reset stats after ramp time completes
512 */
513extern void reset_all_stats(struct thread_data *);
514
515/*
516 * blktrace support
517 */
518#ifdef FIO_HAVE_BLKTRACE
519extern int is_blktrace(const char *, int *);
520extern int load_blktrace(struct thread_data *, const char *, int);
521#endif
522
523/*
524 * Latency target helpers
525 */
526extern void lat_target_check(struct thread_data *);
527extern void lat_target_init(struct thread_data *);
528extern void lat_target_reset(struct thread_data *);
529
530#define for_each_td(td, i) \
531 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
532#define for_each_file(td, f, i) \
533 if ((td)->files_index) \
534 for ((i) = 0, (f) = (td)->files[0]; \
535 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
536 (i)++)
537
538#define fio_assert(td, cond) do { \
539 if (!(cond)) { \
540 int *__foo = NULL; \
541 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
542 td_set_runstate((td), TD_EXITED); \
543 (td)->error = EFAULT; \
544 *__foo = 0; \
545 } \
546} while (0)
547
548static inline int fio_fill_issue_time(struct thread_data *td)
549{
550 if (td->o.read_iolog_file ||
551 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
552 return 1;
553
554 return 0;
555}
556
557static inline int __should_check_rate(struct thread_data *td,
558 enum fio_ddir ddir)
559{
560 struct thread_options *o = &td->o;
561
562 /*
563 * If some rate setting was given, we need to check it
564 */
565 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
566 o->rate_iops_min[ddir])
567 return 1;
568
569 return 0;
570}
571
572static inline int should_check_rate(struct thread_data *td,
573 uint64_t *bytes_done)
574{
575 int ret = 0;
576
577 if (bytes_done[DDIR_READ])
578 ret |= __should_check_rate(td, DDIR_READ);
579 if (bytes_done[DDIR_WRITE])
580 ret |= __should_check_rate(td, DDIR_WRITE);
581 if (bytes_done[DDIR_TRIM])
582 ret |= __should_check_rate(td, DDIR_TRIM);
583
584 return ret;
585}
586
587static inline unsigned int td_max_bs(struct thread_data *td)
588{
589 unsigned int max_bs;
590
591 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
592 return max(td->o.max_bs[DDIR_TRIM], max_bs);
593}
594
595static inline unsigned int td_min_bs(struct thread_data *td)
596{
597 unsigned int min_bs;
598
599 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
600 return min(td->o.min_bs[DDIR_TRIM], min_bs);
601}
602
603static inline int is_power_of_2(unsigned long val)
604{
605 return (val != 0 && ((val & (val - 1)) == 0));
606}
607
608/*
609 * We currently only need to do locking if we have verifier threads
610 * accessing our internal structures too
611 */
612static inline void td_io_u_lock(struct thread_data *td)
613{
614 if (td->o.verify_async)
615 pthread_mutex_lock(&td->io_u_lock);
616}
617
618static inline void td_io_u_unlock(struct thread_data *td)
619{
620 if (td->o.verify_async)
621 pthread_mutex_unlock(&td->io_u_lock);
622}
623
624static inline void td_io_u_free_notify(struct thread_data *td)
625{
626 if (td->o.verify_async)
627 pthread_cond_signal(&td->free_cond);
628}
629
630extern const char *fio_get_arch_string(int);
631extern const char *fio_get_os_string(int);
632
633#ifdef FIO_INTERNAL
634#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
635#endif
636
637enum {
638 FIO_OUTPUT_TERSE = 0,
639 FIO_OUTPUT_JSON,
640 FIO_OUTPUT_NORMAL,
641};
642
643enum {
644 FIO_RAND_DIST_RANDOM = 0,
645 FIO_RAND_DIST_ZIPF,
646 FIO_RAND_DIST_PARETO,
647};
648
649enum {
650 FIO_RAND_GEN_TAUSWORTHE = 0,
651 FIO_RAND_GEN_LFSR,
652};
653
654enum {
655 FIO_CPUS_SHARED = 0,
656 FIO_CPUS_SPLIT,
657};
658
659#endif