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