projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Don't malloc/memcpy ioengine_ops on td initialization
[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 "oslib/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 TD_F_REGROW_LOGS = 1U << 14,
83};
84
85enum {
86 FIO_RAND_BS_OFF = 0,
87 FIO_RAND_VER_OFF,
88 FIO_RAND_MIX_OFF,
89 FIO_RAND_FILE_OFF,
90 FIO_RAND_BLOCK_OFF,
91 FIO_RAND_FILE_SIZE_OFF,
92 FIO_RAND_TRIM_OFF,
93 FIO_RAND_BUF_OFF,
94 FIO_RAND_SEQ_RAND_READ_OFF,
95 FIO_RAND_SEQ_RAND_WRITE_OFF,
96 FIO_RAND_SEQ_RAND_TRIM_OFF,
97 FIO_RAND_START_DELAY,
98 FIO_DEDUPE_OFF,
99 FIO_RAND_POISSON_OFF,
100 FIO_RAND_ZONE_OFF,
101 FIO_RAND_NR_OFFS,
102};
103
104enum {
105 IO_MODE_INLINE = 0,
106 IO_MODE_OFFLOAD = 1,
107
108 RATE_PROCESS_LINEAR = 0,
109 RATE_PROCESS_POISSON = 1,
110};
111
112/*
113 * Per-thread/process specific data. Only used for the network client
114 * for now.
115 */
116struct sk_out;
117void sk_out_assign(struct sk_out *);
118void sk_out_drop(void);
119
120struct zone_split_index {
121 uint8_t size_perc;
122 uint8_t size_perc_prev;
123};
124
125/*
126 * This describes a single thread/process executing a fio job.
127 */
128struct thread_data {
129 struct thread_options o;
130 struct flist_head opt_list;
131 unsigned long flags;
132 void *eo;
133 char verror[FIO_VERROR_SIZE];
134 pthread_t thread;
135 unsigned int thread_number;
136 unsigned int subjob_number;
137 unsigned int groupid;
138 struct thread_stat ts;
139
140 int client_type;
141
142 struct io_log *slat_log;
143 struct io_log *clat_log;
144 struct io_log *lat_log;
145 struct io_log *bw_log;
146 struct io_log *iops_log;
147
148 struct workqueue log_compress_wq;
149
150 struct thread_data *parent;
151
152 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
153 struct timeval bw_sample_time;
154
155 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
156 struct timeval iops_sample_time;
157
158 volatile int update_rusage;
159 struct fio_mutex *rusage_sem;
160 struct rusage ru_start;
161 struct rusage ru_end;
162
163 struct fio_file **files;
164 unsigned char *file_locks;
165 unsigned int files_size;
166 unsigned int files_index;
167 unsigned int nr_open_files;
168 unsigned int nr_done_files;
169 unsigned int nr_normal_files;
170 union {
171 unsigned int next_file;
172 struct frand_state next_file_state;
173 };
174 union {
175 struct zipf_state next_file_zipf;
176 struct gauss_state next_file_gauss;
177 };
178 union {
179 double zipf_theta;
180 double pareto_h;
181 double gauss_dev;
182 };
183 int error;
184 int sig;
185 int done;
186 int stop_io;
187 pid_t pid;
188 char *orig_buffer;
189 size_t orig_buffer_size;
190 volatile int terminate;
191 volatile int runstate;
192 unsigned int last_was_sync;
193 enum fio_ddir last_ddir;
194
195 int mmapfd;
196
197 void *iolog_buf;
198 FILE *iolog_f;
199
200 char *sysfs_root;
201
202 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
203
204 struct frand_state bsrange_state;
205 struct frand_state verify_state;
206 struct frand_state trim_state;
207 struct frand_state delay_state;
208
209 struct frand_state buf_state;
210 struct frand_state buf_state_prev;
211 struct frand_state dedupe_state;
212 struct frand_state zone_state;
213
214 struct zone_split_index **zone_state_index;
215
216 unsigned int verify_batch;
217 unsigned int trim_batch;
218
219 struct thread_io_list *vstate;
220
221 int shm_id;
222
223 /*
224 * IO engine hooks, contains everything needed to submit an io_u
225 * to any of the available IO engines.
226 */
227 struct ioengine_ops *io_ops;
228
229 /*
230 * IO engine private data and dlhandle.
231 */
232 void *io_ops_data;
233 void *io_ops_dlhandle;
234
235 /*
236 * Queue depth of io_u's that fio MIGHT do
237 */
238 unsigned int cur_depth;
239
240 /*
241 * io_u's about to be committed
242 */
243 unsigned int io_u_queued;
244
245 /*
246 * io_u's submitted but not completed yet
247 */
248 unsigned int io_u_in_flight;
249
250 /*
251 * List of free and busy io_u's
252 */
253 struct io_u_ring io_u_requeues;
254 struct io_u_queue io_u_freelist;
255 struct io_u_queue io_u_all;
256 pthread_mutex_t io_u_lock;
257 pthread_cond_t free_cond;
258
259 /*
260 * async verify offload
261 */
262 struct flist_head verify_list;
263 pthread_t *verify_threads;
264 unsigned int nr_verify_threads;
265 pthread_cond_t verify_cond;
266 int verify_thread_exit;
267
268 /*
269 * Rate state
270 */
271 uint64_t rate_bps[DDIR_RWDIR_CNT];
272 unsigned long rate_next_io_time[DDIR_RWDIR_CNT];
273 unsigned long rate_bytes[DDIR_RWDIR_CNT];
274 unsigned long rate_blocks[DDIR_RWDIR_CNT];
275 unsigned long rate_io_issue_bytes[DDIR_RWDIR_CNT];
276 struct timeval lastrate[DDIR_RWDIR_CNT];
277 int64_t last_usec;
278 struct frand_state poisson_state;
279
280 /*
281 * Enforced rate submission/completion workqueue
282 */
283 struct workqueue io_wq;
284
285 uint64_t total_io_size;
286 uint64_t fill_device_size;
287
288 /*
289 * Issue side
290 */
291 uint64_t io_issues[DDIR_RWDIR_CNT];
292 uint64_t io_issue_bytes[DDIR_RWDIR_CNT];
293 uint64_t loops;
294
295 /*
296 * Completions
297 */
298 uint64_t io_blocks[DDIR_RWDIR_CNT];
299 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
300 uint64_t io_bytes[DDIR_RWDIR_CNT];
301 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
302 uint64_t io_skip_bytes;
303 uint64_t zone_bytes;
304 struct fio_mutex *mutex;
305 uint64_t bytes_done[DDIR_RWDIR_CNT];
306
307 /*
308 * State for random io, a bitmap of blocks done vs not done
309 */
310 struct frand_state random_state;
311
312 struct timeval start; /* start of this loop */
313 struct timeval epoch; /* time job was started */
314 struct timeval last_issue;
315 long time_offset;
316 struct timeval tv_cache;
317 struct timeval terminate_time;
318 unsigned int tv_cache_nr;
319 unsigned int tv_cache_mask;
320 unsigned int ramp_time_over;
321
322 /*
323 * Time since last latency_window was started
324 */
325 struct timeval latency_ts;
326 unsigned int latency_qd;
327 unsigned int latency_qd_high;
328 unsigned int latency_qd_low;
329 unsigned int latency_failed;
330 uint64_t latency_ios;
331 int latency_end_run;
332
333 /*
334 * read/write mixed workload state
335 */
336 struct frand_state rwmix_state;
337 unsigned long rwmix_issues;
338 enum fio_ddir rwmix_ddir;
339 unsigned int ddir_seq_nr;
340
341 /*
342 * rand/seq mixed workload state
343 */
344 struct frand_state seq_rand_state[DDIR_RWDIR_CNT];
345
346 /*
347 * IO history logs for verification. We use a tree for sorting,
348 * if we are overwriting. Otherwise just use a fifo.
349 */
350 struct rb_root io_hist_tree;
351 struct flist_head io_hist_list;
352 unsigned long io_hist_len;
353
354 /*
355 * For IO replaying
356 */
357 struct flist_head io_log_list;
358
359 /*
360 * For tracking/handling discards
361 */
362 struct flist_head trim_list;
363 unsigned long trim_entries;
364
365 struct flist_head next_rand_list;
366
367 /*
368 * for fileservice, how often to switch to a new file
369 */
370 unsigned int file_service_nr;
371 unsigned int file_service_left;
372 struct fio_file *file_service_file;
373
374 unsigned int sync_file_range_nr;
375
376 /*
377 * For generating file sizes
378 */
379 struct frand_state file_size_state;
380
381 /*
382 * Error counts
383 */
384 unsigned int total_err_count;
385 int first_error;
386
387 struct fio_flow *flow;
388
389 /*
390 * Can be overloaded by profiles
391 */
392 struct prof_io_ops prof_io_ops;
393 void *prof_data;
394
395 void *pinned_mem;
396};
397
398/*
399 * when should interactive ETA output be generated
400 */
401enum {
402 FIO_ETA_AUTO,
403 FIO_ETA_ALWAYS,
404 FIO_ETA_NEVER,
405};
406
407#define __td_verror(td, err, msg, func) \
408 do { \
409 unsigned int ____e = (err); \
410 if ((td)->error) \
411 break; \
412 (td)->error = ____e; \
413 if (!(td)->first_error) \
414 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
415 } while (0)
416
417
418#define td_clear_error(td) do { \
419 (td)->error = 0; \
420 if ((td)->parent) \
421 (td)->parent->error = 0; \
422} while (0)
423
424#define td_verror(td, err, func) do { \
425 __td_verror((td), (err), strerror((err)), (func)); \
426 if ((td)->parent) \
427 __td_verror((td)->parent, (err), strerror((err)), (func)); \
428} while (0)
429
430#define td_vmsg(td, err, msg, func) do { \
431 __td_verror((td), (err), (msg), (func)); \
432 if ((td)->parent) \
433 __td_verror((td)->parent, (err), (msg), (func)); \
434} while (0)
435
436#define __fio_stringify_1(x) #x
437#define __fio_stringify(x) __fio_stringify_1(x)
438
439extern int exitall_on_terminate;
440extern unsigned int thread_number;
441extern unsigned int stat_number;
442extern int shm_id;
443extern int groupid;
444extern int output_format;
445extern int append_terse_output;
446extern int temp_stall_ts;
447extern uintptr_t page_mask, page_size;
448extern int read_only;
449extern int eta_print;
450extern int eta_new_line;
451extern unsigned long done_secs;
452extern char *job_section;
453extern int fio_gtod_offload;
454extern int fio_gtod_cpu;
455extern enum fio_cs fio_clock_source;
456extern int fio_clock_source_set;
457extern int warnings_fatal;
458extern int terse_version;
459extern int is_backend;
460extern int nr_clients;
461extern int log_syslog;
462extern int status_interval;
463extern const char fio_version_string[];
464extern char *trigger_file;
465extern char *trigger_cmd;
466extern char *trigger_remote_cmd;
467extern long long trigger_timeout;
468extern char *aux_path;
469
470extern struct thread_data *threads;
471
472static inline void fio_ro_check(const struct thread_data *td, struct io_u *io_u)
473{
474 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
475}
476
477#define REAL_MAX_JOBS 2048
478
479static inline int should_fsync(struct thread_data *td)
480{
481 if (td->last_was_sync)
482 return 0;
483 if (td_write(td) || td_rw(td) || td->o.override_sync)
484 return 1;
485
486 return 0;
487}
488
489/*
490 * Init/option functions
491 */
492extern int __must_check fio_init_options(void);
493extern int __must_check parse_options(int, char **);
494extern int parse_jobs_ini(char *, int, int, int);
495extern int parse_cmd_line(int, char **, int);
496extern int fio_backend(struct sk_out *);
497extern void reset_fio_state(void);
498extern void clear_io_state(struct thread_data *, int);
499extern int fio_options_parse(struct thread_data *, char **, int);
500extern void fio_keywords_init(void);
501extern void fio_keywords_exit(void);
502extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
503extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
504extern void fio_fill_default_options(struct thread_data *);
505extern int fio_show_option_help(const char *);
506extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
507extern void fio_options_dup_and_init(struct option *);
508extern void fio_options_mem_dupe(struct thread_data *);
509extern void options_mem_dupe(void *data, struct fio_option *options);
510extern void td_fill_rand_seeds(struct thread_data *);
511extern void td_fill_verify_state_seed(struct thread_data *);
512extern void add_job_opts(const char **, int);
513extern char *num2str(uint64_t, int, int, int, int);
514extern int ioengine_load(struct thread_data *);
515extern int parse_dryrun(void);
516extern int fio_running_or_pending_io_threads(void);
517extern int fio_set_fd_nonblocking(int, const char *);
518extern void sig_show_status(int sig);
519extern struct thread_data *get_global_options(void);
520
521extern uintptr_t page_mask;
522extern uintptr_t page_size;
523extern int initialize_fio(char *envp[]);
524extern void deinitialize_fio(void);
525
526#define FIO_GETOPT_JOB 0x89000000
527#define FIO_GETOPT_IOENGINE 0x98000000
528#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
529
530/*
531 * ETA/status stuff
532 */
533extern void print_thread_status(void);
534extern void print_status_init(int);
535extern char *fio_uint_to_kmg(unsigned int val);
536
537/*
538 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
539 * will never back again. It may cycle between running/verififying/fsyncing.
540 * Once the thread reaches TD_EXITED, it is just waiting for the core to
541 * reap it.
542 */
543enum {
544 TD_NOT_CREATED = 0,
545 TD_CREATED,
546 TD_INITIALIZED,
547 TD_RAMP,
548 TD_SETTING_UP,
549 TD_RUNNING,
550 TD_PRE_READING,
551 TD_VERIFYING,
552 TD_FSYNCING,
553 TD_FINISHING,
554 TD_EXITED,
555 TD_REAPED,
556 TD_LAST,
557};
558
559extern void td_set_runstate(struct thread_data *, int);
560extern int td_bump_runstate(struct thread_data *, int);
561extern void td_restore_runstate(struct thread_data *, int);
562extern const char *runstate_to_name(int runstate);
563
564/*
565 * Allow 60 seconds for a job to quit on its own, otherwise reap with
566 * a vengeance.
567 */
568#define FIO_REAP_TIMEOUT 60
569
570#define TERMINATE_ALL (-1U)
571extern void fio_terminate_threads(unsigned int);
572extern void fio_mark_td_terminate(struct thread_data *);
573
574/*
575 * Memory helpers
576 */
577extern int __must_check fio_pin_memory(struct thread_data *);
578extern void fio_unpin_memory(struct thread_data *);
579extern int __must_check allocate_io_mem(struct thread_data *);
580extern void free_io_mem(struct thread_data *);
581extern void free_threads_shm(void);
582
583/*
584 * Reset stats after ramp time completes
585 */
586extern void reset_all_stats(struct thread_data *);
587
588/*
589 * blktrace support
590 */
591#ifdef FIO_HAVE_BLKTRACE
592extern int is_blktrace(const char *, int *);
593extern int load_blktrace(struct thread_data *, const char *, int);
594#endif
595
596extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
597 enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
598 struct timeval *comp_time);
599
600/*
601 * Latency target helpers
602 */
603extern void lat_target_check(struct thread_data *);
604extern void lat_target_init(struct thread_data *);
605extern void lat_target_reset(struct thread_data *);
606
607#define for_each_td(td, i) \
608 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
609#define for_each_file(td, f, i) \
610 if ((td)->files_index) \
611 for ((i) = 0, (f) = (td)->files[0]; \
612 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
613 (i)++)
614
615#define fio_assert(td, cond) do { \
616 if (!(cond)) { \
617 int *__foo = NULL; \
618 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
619 td_set_runstate((td), TD_EXITED); \
620 (td)->error = EFAULT; \
621 *__foo = 0; \
622 } \
623} while (0)
624
625static inline int fio_fill_issue_time(struct thread_data *td)
626{
627 if (td->o.read_iolog_file ||
628 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
629 return 1;
630
631 return 0;
632}
633
634static inline int __should_check_rate(struct thread_data *td,
635 enum fio_ddir ddir)
636{
637 struct thread_options *o = &td->o;
638
639 /*
640 * If some rate setting was given, we need to check it
641 */
642 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
643 o->rate_iops_min[ddir])
644 return 1;
645
646 return 0;
647}
648
649static inline int should_check_rate(struct thread_data *td)
650{
651 int ret = 0;
652
653 if (td->bytes_done[DDIR_READ])
654 ret |= __should_check_rate(td, DDIR_READ);
655 if (td->bytes_done[DDIR_WRITE])
656 ret |= __should_check_rate(td, DDIR_WRITE);
657 if (td->bytes_done[DDIR_TRIM])
658 ret |= __should_check_rate(td, DDIR_TRIM);
659
660 return ret;
661}
662
663static inline unsigned int td_max_bs(struct thread_data *td)
664{
665 unsigned int max_bs;
666
667 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
668 return max(td->o.max_bs[DDIR_TRIM], max_bs);
669}
670
671static inline unsigned int td_min_bs(struct thread_data *td)
672{
673 unsigned int min_bs;
674
675 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
676 return min(td->o.min_bs[DDIR_TRIM], min_bs);
677}
678
679static inline int td_async_processing(struct thread_data *td)
680{
681 return (td->flags & TD_F_NEED_LOCK) != 0;
682}
683
684/*
685 * We currently only need to do locking if we have verifier threads
686 * accessing our internal structures too
687 */
688static inline void td_io_u_lock(struct thread_data *td)
689{
690 if (td_async_processing(td))
691 pthread_mutex_lock(&td->io_u_lock);
692}
693
694static inline void td_io_u_unlock(struct thread_data *td)
695{
696 if (td_async_processing(td))
697 pthread_mutex_unlock(&td->io_u_lock);
698}
699
700static inline void td_io_u_free_notify(struct thread_data *td)
701{
702 if (td_async_processing(td))
703 pthread_cond_signal(&td->free_cond);
704}
705
706extern const char *fio_get_arch_string(int);
707extern const char *fio_get_os_string(int);
708
709#ifdef FIO_INTERNAL
710#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
711#define FIELD_SIZE(s, f) (sizeof(((typeof(s))0)->f))
712#endif
713
714enum {
715 __FIO_OUTPUT_TERSE = 0,
716 __FIO_OUTPUT_JSON = 1,
717 __FIO_OUTPUT_NORMAL = 2,
718 __FIO_OUTPUT_JSON_PLUS = 3,
719 FIO_OUTPUT_NR = 4,
720
721 FIO_OUTPUT_TERSE = 1U << __FIO_OUTPUT_TERSE,
722 FIO_OUTPUT_JSON = 1U << __FIO_OUTPUT_JSON,
723 FIO_OUTPUT_NORMAL = 1U << __FIO_OUTPUT_NORMAL,
724 FIO_OUTPUT_JSON_PLUS = 1U << __FIO_OUTPUT_JSON_PLUS,
725};
726
727enum {
728 FIO_RAND_DIST_RANDOM = 0,
729 FIO_RAND_DIST_ZIPF,
730 FIO_RAND_DIST_PARETO,
731 FIO_RAND_DIST_GAUSS,
732 FIO_RAND_DIST_ZONED,
733};
734
735#define FIO_DEF_ZIPF 1.1
736#define FIO_DEF_PARETO 0.2
737
738enum {
739 FIO_RAND_GEN_TAUSWORTHE = 0,
740 FIO_RAND_GEN_LFSR,
741 FIO_RAND_GEN_TAUSWORTHE64,
742};
743
744enum {
745 FIO_CPUS_SHARED = 0,
746 FIO_CPUS_SPLIT,
747};
748
749extern void exec_trigger(const char *);
750extern void check_trigger_file(void);
751
752#endif
fio - Flexible IO Tester