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