projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
server: ensure each connection sets up its own sk_out
[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_POISSON_OFF,
99 FIO_RAND_NR_OFFS,
100};
101
102enum {
103 IO_MODE_INLINE = 0,
104 IO_MODE_OFFLOAD = 1,
105
106 RATE_PROCESS_LINEAR = 0,
107 RATE_PROCESS_POISSON = 1,
108};
109
110/*
111 * Per-thread/process specific data. Only used for the network client
112 * for now.
113 */
114struct sk_out;
115void sk_out_assign(struct sk_out *);
116void sk_out_drop(void);
117
118/*
119 * This describes a single thread/process executing a fio job.
120 */
121struct thread_data {
122 struct thread_options o;
123 unsigned long flags;
124 void *eo;
125 char verror[FIO_VERROR_SIZE];
126 pthread_t thread;
127 unsigned int thread_number;
128 unsigned int subjob_number;
129 unsigned int groupid;
130 struct thread_stat ts;
131
132 int client_type;
133
134 struct io_log *slat_log;
135 struct io_log *clat_log;
136 struct io_log *lat_log;
137 struct io_log *bw_log;
138 struct io_log *iops_log;
139
140 struct workqueue log_compress_wq;
141
142 struct thread_data *parent;
143
144 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
145 struct timeval bw_sample_time;
146
147 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
148 struct timeval iops_sample_time;
149
150 /*
151 * Tracks the last iodepth number of completed writes, if data
152 * verification is enabled
153 */
154 uint64_t *last_write_comp;
155 unsigned int last_write_idx;
156
157 volatile int update_rusage;
158 struct fio_mutex *rusage_sem;
159 struct rusage ru_start;
160 struct rusage ru_end;
161
162 struct fio_file **files;
163 unsigned char *file_locks;
164 unsigned int files_size;
165 unsigned int files_index;
166 unsigned int nr_open_files;
167 unsigned int nr_done_files;
168 unsigned int nr_normal_files;
169 union {
170 unsigned int next_file;
171 struct frand_state next_file_state;
172 };
173 int error;
174 int sig;
175 int done;
176 int stop_io;
177 pid_t pid;
178 char *orig_buffer;
179 size_t orig_buffer_size;
180 volatile int terminate;
181 volatile int runstate;
182 unsigned int last_was_sync;
183 enum fio_ddir last_ddir;
184
185 int mmapfd;
186
187 void *iolog_buf;
188 FILE *iolog_f;
189
190 char *sysfs_root;
191
192 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
193
194 struct frand_state bsrange_state;
195 struct frand_state verify_state;
196 struct frand_state trim_state;
197 struct frand_state delay_state;
198
199 struct frand_state buf_state;
200 struct frand_state buf_state_prev;
201 struct frand_state dedupe_state;
202
203 unsigned int verify_batch;
204 unsigned int trim_batch;
205
206 struct thread_io_list *vstate;
207
208 int shm_id;
209
210 /*
211 * IO engine hooks, contains everything needed to submit an io_u
212 * to any of the available IO engines.
213 */
214 struct ioengine_ops *io_ops;
215
216 /*
217 * Queue depth of io_u's that fio MIGHT do
218 */
219 unsigned int cur_depth;
220
221 /*
222 * io_u's about to be committed
223 */
224 unsigned int io_u_queued;
225
226 /*
227 * io_u's submitted but not completed yet
228 */
229 unsigned int io_u_in_flight;
230
231 /*
232 * List of free and busy io_u's
233 */
234 struct io_u_ring io_u_requeues;
235 struct io_u_queue io_u_freelist;
236 struct io_u_queue io_u_all;
237 pthread_mutex_t io_u_lock;
238 pthread_cond_t free_cond;
239
240 /*
241 * async verify offload
242 */
243 struct flist_head verify_list;
244 pthread_t *verify_threads;
245 unsigned int nr_verify_threads;
246 pthread_cond_t verify_cond;
247 int verify_thread_exit;
248
249 /*
250 * Rate state
251 */
252 uint64_t rate_bps[DDIR_RWDIR_CNT];
253 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
254 unsigned long rate_bytes[DDIR_RWDIR_CNT];
255 unsigned long rate_blocks[DDIR_RWDIR_CNT];
256 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
257 struct timeval lastrate[DDIR_RWDIR_CNT];
258 int64_t last_usec;
259 struct frand_state poisson_state;
260
261 /*
262 * Enforced rate submission/completion workqueue
263 */
264 struct workqueue io_wq;
265
266 uint64_t total_io_size;
267 uint64_t fill_device_size;
268
269 /*
270 * Issue side
271 */
272 uint64_t io_issues[DDIR_RWDIR_CNT];
273 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
274 uint64_t loops;
275
276 /*
277 * Completions
278 */
279 uint64_t io_blocks[DDIR_RWDIR_CNT];
280 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
281 uint64_t io_bytes[DDIR_RWDIR_CNT];
282 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
283 uint64_t io_skip_bytes;
284 uint64_t zone_bytes;
285 struct fio_mutex *mutex;
286 uint64_t bytes_done[DDIR_RWDIR_CNT];
287
288 /*
289 * State for random io, a bitmap of blocks done vs not done
290 */
291 struct frand_state random_state;
292
293 struct timeval start; /* start of this loop */
294 struct timeval epoch; /* time job was started */
295 struct timeval last_issue;
296 long time_offset;
297 struct timeval tv_cache;
298 struct timeval terminate_time;
299 unsigned int tv_cache_nr;
300 unsigned int tv_cache_mask;
301 unsigned int ramp_time_over;
302
303 /*
304 * Time since last latency_window was started
305 */
306 struct timeval latency_ts;
307 unsigned int latency_qd;
308 unsigned int latency_qd_high;
309 unsigned int latency_qd_low;
310 unsigned int latency_failed;
311 uint64_t latency_ios;
312 int latency_end_run;
313
314 /*
315 * read/write mixed workload state
316 */
317 struct frand_state rwmix_state;
318 unsigned long rwmix_issues;
319 enum fio_ddir rwmix_ddir;
320 unsigned int ddir_seq_nr;
321
322 /*
323 * rand/seq mixed workload state
324 */
325 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
326
327 /*
328 * IO history logs for verification. We use a tree for sorting,
329 * if we are overwriting. Otherwise just use a fifo.
330 */
331 struct rb_root io_hist_tree;
332 struct flist_head io_hist_list;
333 unsigned long io_hist_len;
334
335 /*
336 * For IO replaying
337 */
338 struct flist_head io_log_list;
339
340 /*
341 * For tracking/handling discards
342 */
343 struct flist_head trim_list;
344 unsigned long trim_entries;
345
346 struct flist_head next_rand_list;
347
348 /*
349 * for fileservice, how often to switch to a new file
350 */
351 unsigned int file_service_nr;
352 unsigned int file_service_left;
353 struct fio_file *file_service_file;
354
355 unsigned int sync_file_range_nr;
356
357 /*
358 * For generating file sizes
359 */
360 struct frand_state file_size_state;
361
362 /*
363 * Error counts
364 */
365 unsigned int total_err_count;
366 int first_error;
367
368 struct fio_flow *flow;
369
370 /*
371 * Can be overloaded by profiles
372 */
373 struct prof_io_ops prof_io_ops;
374 void *prof_data;
375
376 void *pinned_mem;
377};
378
379/*
380 * when should interactive ETA output be generated
381 */
382enum {
383 FIO_ETA_AUTO,
384 FIO_ETA_ALWAYS,
385 FIO_ETA_NEVER,
386};
387
388#define __td_verror(td, err, msg, func) \
389 do { \
390 unsigned int ____e = (err); \
391 if ((td)->error) \
392 break; \
393 (td)->error = ____e; \
394 if (!(td)->first_error) \
395 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
396 } while (0)
397
398
399#define td_clear_error(td) do { \
400 (td)->error = 0; \
401 if ((td)->parent) \
402 (td)->parent->error = 0; \
403} while (0)
404
405#define td_verror(td, err, func) do { \
406 __td_verror((td), (err), strerror((err)), (func)); \
407 if ((td)->parent) \
408 __td_verror((td)->parent, (err), strerror((err)), (func)); \
409} while (0)
410
411#define td_vmsg(td, err, msg, func) do { \
412 __td_verror((td), (err), (msg), (func)); \
413 if ((td)->parent) \
414 __td_verror((td)->parent, (err), (msg), (func)); \
415} while (0)
416
417#define __fio_stringify_1(x) #x
418#define __fio_stringify(x) __fio_stringify_1(x)
419
420extern int exitall_on_terminate;
421extern unsigned int thread_number;
422extern unsigned int stat_number;
423extern int shm_id;
424extern int groupid;
425extern int output_format;
426extern int append_terse_output;
427extern int temp_stall_ts;
428extern uintptr_t page_mask, page_size;
429extern int read_only;
430extern int eta_print;
431extern int eta_new_line;
432extern unsigned long done_secs;
433extern char *job_section;
434extern int fio_gtod_offload;
435extern int fio_gtod_cpu;
436extern enum fio_cs fio_clock_source;
437extern int fio_clock_source_set;
438extern int warnings_fatal;
439extern int terse_version;
440extern int is_backend;
441extern int nr_clients;
442extern int log_syslog;
443extern int status_interval;
444extern const char fio_version_string[];
445extern int helper_do_stat;
446extern pthread_cond_t helper_cond;
447extern char *trigger_file;
448extern char *trigger_cmd;
449extern char *trigger_remote_cmd;
450extern long long trigger_timeout;
451extern char *aux_path;
452
453extern struct thread_data *threads;
454
455static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
456{
457 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
458}
459
460#define REAL_MAX_JOBS 2048
461
462static inline int should_fsync(struct thread_data *td)
463{
464 if (td->last_was_sync)
465 return 0;
466 if (td_write(td) || td_rw(td) || td->o.override_sync)
467 return 1;
468
469 return 0;
470}
471
472/*
473 * Init/option functions
474 */
475extern int __must_check fio_init_options(void);
476extern int __must_check parse_options(int, char **);
477extern int parse_jobs_ini(char *, int, int, int);
478extern int parse_cmd_line(int, char **, int);
479extern int fio_backend(struct sk_out *);
480extern void reset_fio_state(void);
481extern void clear_io_state(struct thread_data *, int);
482extern int fio_options_parse(struct thread_data *, char **, int, int);
483extern void fio_keywords_init(void);
484extern void fio_keywords_exit(void);
485extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
486extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
487extern void fio_fill_default_options(struct thread_data *);
488extern int fio_show_option_help(const char *);
489extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
490extern void fio_options_dup_and_init(struct option *);
491extern void fio_options_mem_dupe(struct thread_data *);
492extern void options_mem_dupe(void *data, struct fio_option *options);
493extern void td_fill_rand_seeds(struct thread_data *);
494extern void add_job_opts(const char **, int);
495extern char *num2str(uint64_t, int, int, int, int);
496extern int ioengine_load(struct thread_data *);
497extern int parse_dryrun(void);
498extern int fio_running_or_pending_io_threads(void);
499extern int fio_set_fd_nonblocking(int, const char *);
500extern void sig_show_status(int sig);
501
502extern uintptr_t page_mask;
503extern uintptr_t page_size;
504extern int initialize_fio(char *envp[]);
505extern void deinitialize_fio(void);
506
507#define FIO_GETOPT_JOB 0x89000000
508#define FIO_GETOPT_IOENGINE 0x98000000
509#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
510
511/*
512 * ETA/status stuff
513 */
514extern void print_thread_status(void);
515extern void print_status_init(int);
516extern char *fio_uint_to_kmg(unsigned int val);
517
518/*
519 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
520 * will never back again. It may cycle between running/verififying/fsyncing.
521 * Once the thread reaches TD_EXITED, it is just waiting for the core to
522 * reap it.
523 */
524enum {
525 TD_NOT_CREATED = 0,
526 TD_CREATED,
527 TD_INITIALIZED,
528 TD_RAMP,
529 TD_SETTING_UP,
530 TD_RUNNING,
531 TD_PRE_READING,
532 TD_VERIFYING,
533 TD_FSYNCING,
534 TD_FINISHING,
535 TD_EXITED,
536 TD_REAPED,
537 TD_LAST,
538};
539
540extern void td_set_runstate(struct thread_data *, int);
541extern int td_bump_runstate(struct thread_data *, int);
542extern void td_restore_runstate(struct thread_data *, int);
543
544/*
545 * Allow 60 seconds for a job to quit on its own, otherwise reap with
546 * a vengeance.
547 */
548#define FIO_REAP_TIMEOUT 60
549
550#define TERMINATE_ALL (-1U)
551extern void fio_terminate_threads(unsigned int);
552extern void fio_mark_td_terminate(struct thread_data *);
553
554/*
555 * Memory helpers
556 */
557extern int __must_check fio_pin_memory(struct thread_data *);
558extern void fio_unpin_memory(struct thread_data *);
559extern int __must_check allocate_io_mem(struct thread_data *);
560extern void free_io_mem(struct thread_data *);
561extern void free_threads_shm(void);
562
563/*
564 * Reset stats after ramp time completes
565 */
566extern void reset_all_stats(struct thread_data *);
567
568/*
569 * blktrace support
570 */
571#ifdef FIO_HAVE_BLKTRACE
572extern int is_blktrace(const char *, int *);
573extern int load_blktrace(struct thread_data *, const char *, int);
574#endif
575
576extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
577 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
578 struct timeval *comp_time);
579
580/*
581 * Latency target helpers
582 */
583extern void lat_target_check(struct thread_data *);
584extern void lat_target_init(struct thread_data *);
585extern void lat_target_reset(struct thread_data *);
586
587#define for_each_td(td, i) \
588 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
589#define for_each_file(td, f, i) \
590 if ((td)->files_index) \
591 for ((i) = 0, (f) = (td)->files[0]; \
592 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
593 (i)++)
594
595#define fio_assert(td, cond) do { \
596 if (!(cond)) { \
597 int *__foo = NULL; \
598 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
599 td_set_runstate((td), TD_EXITED); \
600 (td)->error = EFAULT; \
601 *__foo = 0; \
602 } \
603} while (0)
604
605static inline int fio_fill_issue_time(struct thread_data *td)
606{
607 if (td->o.read_iolog_file ||
608 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
609 return 1;
610
611 return 0;
612}
613
614static inline int __should_check_rate(struct thread_data *td,
615 enum fio_ddir ddir)
616{
617 struct thread_options *o = &td->o;
618
619 /*
620 * If some rate setting was given, we need to check it
621 */
622 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
623 o->rate_iops_min[ddir])
624 return 1;
625
626 return 0;
627}
628
629static inline int should_check_rate(struct thread_data *td)
630{
631 int ret = 0;
632
633 if (td->bytes_done[DDIR_READ])
634 ret |= __should_check_rate(td, DDIR_READ);
635 if (td->bytes_done[DDIR_WRITE])
636 ret |= __should_check_rate(td, DDIR_WRITE);
637 if (td->bytes_done[DDIR_TRIM])
638 ret |= __should_check_rate(td, DDIR_TRIM);
639
640 return ret;
641}
642
643static inline unsigned int td_max_bs(struct thread_data *td)
644{
645 unsigned int max_bs;
646
647 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
648 return max(td->o.max_bs[DDIR_TRIM], max_bs);
649}
650
651static inline unsigned int td_min_bs(struct thread_data *td)
652{
653 unsigned int min_bs;
654
655 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
656 return min(td->o.min_bs[DDIR_TRIM], min_bs);
657}
658
659static inline int td_async_processing(struct thread_data *td)
660{
661 return (td->flags & TD_F_NEED_LOCK) != 0;
662}
663
664/*
665 * We currently only need to do locking if we have verifier threads
666 * accessing our internal structures too
667 */
668static inline void td_io_u_lock(struct thread_data *td)
669{
670 if (td_async_processing(td))
671 pthread_mutex_lock(&td->io_u_lock);
672}
673
674static inline void td_io_u_unlock(struct thread_data *td)
675{
676 if (td_async_processing(td))
677 pthread_mutex_unlock(&td->io_u_lock);
678}
679
680static inline void td_io_u_free_notify(struct thread_data *td)
681{
682 if (td_async_processing(td))
683 pthread_cond_signal(&td->free_cond);
684}
685
686extern const char *fio_get_arch_string(int);
687extern const char *fio_get_os_string(int);
688
689#ifdef FIO_INTERNAL
690#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
691#define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
692#endif
693
694enum {
695 __FIO_OUTPUT_TERSE = 0,
696 __FIO_OUTPUT_JSON = 1,
697 __FIO_OUTPUT_NORMAL = 2,
698 __FIO_OUTPUT_JSON_PLUS = 3,
699 FIO_OUTPUT_NR = 4,
700
701 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
702 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
703 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
704 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
705};
706
707enum {
708 FIO_RAND_DIST_RANDOM = 0,
709 FIO_RAND_DIST_ZIPF,
710 FIO_RAND_DIST_PARETO,
711 FIO_RAND_DIST_GAUSS,
712};
713
714#define FIO_DEF_ZIPF 1.1
715#define FIO_DEF_PARETO 0.2
716
717enum {
718 FIO_RAND_GEN_TAUSWORTHE = 0,
719 FIO_RAND_GEN_LFSR,
720 FIO_RAND_GEN_TAUSWORTHE64,
721};
722
723enum {
724 FIO_CPUS_SHARED = 0,
725 FIO_CPUS_SPLIT,
726};
727
728extern void exec_trigger(const char *);
729extern void check_trigger_file(void);
730
731#endif
fio - Flexible IO Tester