projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Cleanup the parallellized log writing
[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 "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};
76
77enum {
78 FIO_RAND_BS_OFF = 0,
79 FIO_RAND_VER_OFF,
80 FIO_RAND_MIX_OFF,
81 FIO_RAND_FILE_OFF,
82 FIO_RAND_BLOCK_OFF,
83 FIO_RAND_FILE_SIZE_OFF,
84 FIO_RAND_TRIM_OFF,
85 FIO_RAND_BUF_OFF,
86 FIO_RAND_SEQ_RAND_READ_OFF,
87 FIO_RAND_SEQ_RAND_WRITE_OFF,
88 FIO_RAND_SEQ_RAND_TRIM_OFF,
89 FIO_RAND_START_DELAY,
90 FIO_RAND_NR_OFFS,
91};
92
93/*
94 * This describes a single thread/process executing a fio job.
95 */
96struct thread_data {
97 struct thread_options o;
98 unsigned long flags;
99 void *eo;
100 char verror[FIO_VERROR_SIZE];
101 pthread_t thread;
102 unsigned int thread_number;
103 unsigned int groupid;
104 struct thread_stat ts;
105
106 int client_type;
107
108 struct io_log *slat_log;
109 struct io_log *clat_log;
110 struct io_log *lat_log;
111 struct io_log *bw_log;
112 struct io_log *iops_log;
113
114 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
115 struct timeval bw_sample_time;
116
117 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
118 struct timeval iops_sample_time;
119
120 volatile int update_rusage;
121 struct fio_mutex *rusage_sem;
122 struct rusage ru_start;
123 struct rusage ru_end;
124
125 struct fio_file **files;
126 unsigned char *file_locks;
127 unsigned int files_size;
128 unsigned int files_index;
129 unsigned int nr_open_files;
130 unsigned int nr_done_files;
131 unsigned int nr_normal_files;
132 union {
133 unsigned int next_file;
134 os_random_state_t next_file_state;
135 struct frand_state __next_file_state;
136 };
137 int error;
138 int sig;
139 int done;
140 pid_t pid;
141 char *orig_buffer;
142 size_t orig_buffer_size;
143 volatile int terminate;
144 volatile int runstate;
145 unsigned int last_was_sync;
146 enum fio_ddir last_ddir;
147
148 int mmapfd;
149
150 void *iolog_buf;
151 FILE *iolog_f;
152
153 char *sysfs_root;
154
155 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
156
157 union {
158 os_random_state_t bsrange_state;
159 struct frand_state __bsrange_state;
160 };
161 union {
162 os_random_state_t verify_state;
163 struct frand_state __verify_state;
164 };
165 union {
166 os_random_state_t trim_state;
167 struct frand_state __trim_state;
168 };
169 union {
170 os_random_state_t delay_state;
171 struct frand_state __delay_state;
172 };
173
174 struct frand_state buf_state;
175
176 unsigned int verify_batch;
177 unsigned int trim_batch;
178
179 int shm_id;
180
181 /*
182 * IO engine hooks, contains everything needed to submit an io_u
183 * to any of the available IO engines.
184 */
185 struct ioengine_ops *io_ops;
186
187 /*
188 * Queue depth of io_u's that fio MIGHT do
189 */
190 unsigned int cur_depth;
191
192 /*
193 * io_u's about to be committed
194 */
195 unsigned int io_u_queued;
196
197 /*
198 * io_u's submitted but not completed yet
199 */
200 unsigned int io_u_in_flight;
201
202 /*
203 * List of free and busy io_u's
204 */
205 struct io_u_ring io_u_requeues;
206 struct io_u_queue io_u_freelist;
207 struct io_u_queue io_u_all;
208 pthread_mutex_t io_u_lock;
209 pthread_cond_t free_cond;
210
211 /*
212 * async verify offload
213 */
214 struct flist_head verify_list;
215 pthread_t *verify_threads;
216 unsigned int nr_verify_threads;
217 pthread_cond_t verify_cond;
218 int verify_thread_exit;
219
220 /*
221 * Rate state
222 */
223 uint64_t rate_bps[DDIR_RWDIR_CNT];
224 long rate_pending_usleep[DDIR_RWDIR_CNT];
225 unsigned long rate_bytes[DDIR_RWDIR_CNT];
226 unsigned long rate_blocks[DDIR_RWDIR_CNT];
227 struct timeval lastrate[DDIR_RWDIR_CNT];
228
229 uint64_t total_io_size;
230 uint64_t fill_device_size;
231
232 unsigned long io_issues[DDIR_RWDIR_CNT];
233 uint64_t io_blocks[DDIR_RWDIR_CNT];
234 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
235 uint64_t io_bytes[DDIR_RWDIR_CNT];
236 uint64_t io_skip_bytes;
237 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
238 uint64_t zone_bytes;
239 struct fio_mutex *mutex;
240
241 /*
242 * State for random io, a bitmap of blocks done vs not done
243 */
244 union {
245 os_random_state_t random_state;
246 struct frand_state __random_state;
247 };
248
249 struct timeval start; /* start of this loop */
250 struct timeval epoch; /* time job was started */
251 struct timeval last_issue;
252 struct timeval tv_cache;
253 unsigned int tv_cache_nr;
254 unsigned int tv_cache_mask;
255 unsigned int ramp_time_over;
256
257 /*
258 * Time since last latency_window was started
259 */
260 struct timeval latency_ts;
261 unsigned int latency_qd;
262 unsigned int latency_qd_high;
263 unsigned int latency_qd_low;
264 unsigned int latency_failed;
265 uint64_t latency_ios;
266 int latency_end_run;
267
268 /*
269 * read/write mixed workload state
270 */
271 union {
272 os_random_state_t rwmix_state;
273 struct frand_state __rwmix_state;
274 };
275 unsigned long rwmix_issues;
276 enum fio_ddir rwmix_ddir;
277 unsigned int ddir_seq_nr;
278
279 /*
280 * rand/seq mixed workload state
281 */
282 union {
283 os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
284 struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
285 };
286
287 /*
288 * IO history logs for verification. We use a tree for sorting,
289 * if we are overwriting. Otherwise just use a fifo.
290 */
291 struct rb_root io_hist_tree;
292 struct flist_head io_hist_list;
293 unsigned long io_hist_len;
294
295 /*
296 * For IO replaying
297 */
298 struct flist_head io_log_list;
299
300 /*
301 * For tracking/handling discards
302 */
303 struct flist_head trim_list;
304 unsigned long trim_entries;
305
306 struct flist_head next_rand_list;
307
308 /*
309 * for fileservice, how often to switch to a new file
310 */
311 unsigned int file_service_nr;
312 unsigned int file_service_left;
313 struct fio_file *file_service_file;
314
315 unsigned int sync_file_range_nr;
316
317 /*
318 * For generating file sizes
319 */
320 union {
321 os_random_state_t file_size_state;
322 struct frand_state __file_size_state;
323 };
324
325 /*
326 * Error counts
327 */
328 unsigned int total_err_count;
329 int first_error;
330
331 struct fio_flow *flow;
332
333 /*
334 * Can be overloaded by profiles
335 */
336 struct prof_io_ops prof_io_ops;
337 void *prof_data;
338
339 void *pinned_mem;
340};
341
342/*
343 * when should interactive ETA output be generated
344 */
345enum {
346 FIO_ETA_AUTO,
347 FIO_ETA_ALWAYS,
348 FIO_ETA_NEVER,
349};
350
351#define __td_verror(td, err, msg, func) \
352 do { \
353 int ____e = (err); \
354 if ((td)->error) \
355 break; \
356 (td)->error = ____e; \
357 if (!(td)->first_error) \
358 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
359 } while (0)
360
361
362#define td_clear_error(td) \
363 (td)->error = 0;
364#define td_verror(td, err, func) \
365 __td_verror((td), (err), strerror((err)), (func))
366#define td_vmsg(td, err, msg, func) \
367 __td_verror((td), (err), (msg), (func))
368
369#define __fio_stringify_1(x) #x
370#define __fio_stringify(x) __fio_stringify_1(x)
371
372extern int exitall_on_terminate;
373extern unsigned int thread_number;
374extern unsigned int stat_number;
375extern int shm_id;
376extern int groupid;
377extern int output_format;
378extern int append_terse_output;
379extern int temp_stall_ts;
380extern uintptr_t page_mask, page_size;
381extern int read_only;
382extern int eta_print;
383extern int eta_new_line;
384extern unsigned long done_secs;
385extern char *job_section;
386extern int fio_gtod_offload;
387extern int fio_gtod_cpu;
388extern enum fio_cs fio_clock_source;
389extern int fio_clock_source_set;
390extern int warnings_fatal;
391extern int terse_version;
392extern int is_backend;
393extern int nr_clients;
394extern int log_syslog;
395extern int status_interval;
396extern const char fio_version_string[];
397
398extern struct thread_data *threads;
399
400static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
401{
402 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
403}
404
405#define REAL_MAX_JOBS 2048
406
407static inline int should_fsync(struct thread_data *td)
408{
409 if (td->last_was_sync)
410 return 0;
411 if (td_write(td) || td_rw(td) || td->o.override_sync)
412 return 1;
413
414 return 0;
415}
416
417/*
418 * Init/option functions
419 */
420extern int __must_check fio_init_options(void);
421extern int __must_check parse_options(int, char **);
422extern int parse_jobs_ini(char *, int, int, int);
423extern int parse_cmd_line(int, char **, int);
424extern int fio_backend(void);
425extern void reset_fio_state(void);
426extern void clear_io_state(struct thread_data *);
427extern int fio_options_parse(struct thread_data *, char **, int, int);
428extern void fio_keywords_init(void);
429extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
430extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
431extern void fio_fill_default_options(struct thread_data *);
432extern int fio_show_option_help(const char *);
433extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
434extern void fio_options_dup_and_init(struct option *);
435extern void fio_options_mem_dupe(struct thread_data *);
436extern void options_mem_dupe(void *data, struct fio_option *options);
437extern void td_fill_rand_seeds(struct thread_data *);
438extern void add_job_opts(const char **, int);
439extern char *num2str(unsigned long, int, int, int, int);
440extern int ioengine_load(struct thread_data *);
441extern int parse_dryrun(void);
442
443extern uintptr_t page_mask;
444extern uintptr_t page_size;
445extern int initialize_fio(char *envp[]);
446
447#define FIO_GETOPT_JOB 0x89000000
448#define FIO_GETOPT_IOENGINE 0x98000000
449#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
450
451/*
452 * ETA/status stuff
453 */
454extern void print_thread_status(void);
455extern void print_status_init(int);
456extern char *fio_uint_to_kmg(unsigned int val);
457
458/*
459 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
460 * will never back again. It may cycle between running/verififying/fsyncing.
461 * Once the thread reaches TD_EXITED, it is just waiting for the core to
462 * reap it.
463 */
464enum {
465 TD_NOT_CREATED = 0,
466 TD_CREATED,
467 TD_INITIALIZED,
468 TD_RAMP,
469 TD_SETTING_UP,
470 TD_RUNNING,
471 TD_PRE_READING,
472 TD_VERIFYING,
473 TD_FSYNCING,
474 TD_FINISHING,
475 TD_EXITED,
476 TD_REAPED,
477};
478
479extern void td_set_runstate(struct thread_data *, int);
480extern int td_bump_runstate(struct thread_data *, int);
481extern void td_restore_runstate(struct thread_data *, int);
482
483#define TERMINATE_ALL (-1)
484extern void fio_terminate_threads(int);
485
486/*
487 * Memory helpers
488 */
489extern int __must_check fio_pin_memory(struct thread_data *);
490extern void fio_unpin_memory(struct thread_data *);
491extern int __must_check allocate_io_mem(struct thread_data *);
492extern void free_io_mem(struct thread_data *);
493extern void free_threads_shm(void);
494
495/*
496 * Reset stats after ramp time completes
497 */
498extern void reset_all_stats(struct thread_data *);
499
500/*
501 * blktrace support
502 */
503#ifdef FIO_HAVE_BLKTRACE
504extern int is_blktrace(const char *, int *);
505extern int load_blktrace(struct thread_data *, const char *, int);
506#endif
507
508/*
509 * Latency target helpers
510 */
511extern void lat_target_check(struct thread_data *);
512extern void lat_target_init(struct thread_data *);
513extern void lat_target_reset(struct thread_data *);
514
515#define for_each_td(td, i) \
516 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
517#define for_each_file(td, f, i) \
518 if ((td)->files_index) \
519 for ((i) = 0, (f) = (td)->files[0]; \
520 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
521 (i)++)
522
523#define fio_assert(td, cond) do { \
524 if (!(cond)) { \
525 int *__foo = NULL; \
526 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
527 td_set_runstate((td), TD_EXITED); \
528 (td)->error = EFAULT; \
529 *__foo = 0; \
530 } \
531} while (0)
532
533static inline int fio_fill_issue_time(struct thread_data *td)
534{
535 if (td->o.read_iolog_file ||
536 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
537 return 1;
538
539 return 0;
540}
541
542static inline int __should_check_rate(struct thread_data *td,
543 enum fio_ddir ddir)
544{
545 struct thread_options *o = &td->o;
546
547 /*
548 * If some rate setting was given, we need to check it
549 */
550 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
551 o->rate_iops_min[ddir])
552 return 1;
553
554 return 0;
555}
556
557static inline int should_check_rate(struct thread_data *td,
558 uint64_t *bytes_done)
559{
560 int ret = 0;
561
562 if (bytes_done[DDIR_READ])
563 ret |= __should_check_rate(td, DDIR_READ);
564 if (bytes_done[DDIR_WRITE])
565 ret |= __should_check_rate(td, DDIR_WRITE);
566 if (bytes_done[DDIR_TRIM])
567 ret |= __should_check_rate(td, DDIR_TRIM);
568
569 return ret;
570}
571
572static inline unsigned int td_max_bs(struct thread_data *td)
573{
574 unsigned int max_bs;
575
576 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
577 return max(td->o.max_bs[DDIR_TRIM], max_bs);
578}
579
580static inline unsigned int td_min_bs(struct thread_data *td)
581{
582 unsigned int min_bs;
583
584 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
585 return min(td->o.min_bs[DDIR_TRIM], min_bs);
586}
587
588static inline int is_power_of_2(unsigned int val)
589{
590 return (val != 0 && ((val & (val - 1)) == 0));
591}
592
593/*
594 * We currently only need to do locking if we have verifier threads
595 * accessing our internal structures too
596 */
597static inline void td_io_u_lock(struct thread_data *td)
598{
599 if (td->o.verify_async)
600 pthread_mutex_lock(&td->io_u_lock);
601}
602
603static inline void td_io_u_unlock(struct thread_data *td)
604{
605 if (td->o.verify_async)
606 pthread_mutex_unlock(&td->io_u_lock);
607}
608
609static inline void td_io_u_free_notify(struct thread_data *td)
610{
611 if (td->o.verify_async)
612 pthread_cond_signal(&td->free_cond);
613}
614
615extern const char *fio_get_arch_string(int);
616extern const char *fio_get_os_string(int);
617
618#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
619
620enum {
621 FIO_OUTPUT_TERSE = 0,
622 FIO_OUTPUT_JSON,
623 FIO_OUTPUT_NORMAL,
624};
625
626enum {
627 FIO_RAND_DIST_RANDOM = 0,
628 FIO_RAND_DIST_ZIPF,
629 FIO_RAND_DIST_PARETO,
630};
631
632enum {
633 FIO_RAND_GEN_TAUSWORTHE = 0,
634 FIO_RAND_GEN_LFSR,
635};
636
637enum {
638 FIO_CPUS_SHARED = 0,
639 FIO_CPUS_SPLIT,
640};
641
642#endif
fio - Flexible IO Tester