Add support for runtime log compression
[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
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 TD_F_NOIO = 256,
76 TD_F_COMPRESS_LOG = 512,
77};
78
79enum {
80 FIO_RAND_BS_OFF = 0,
81 FIO_RAND_VER_OFF,
82 FIO_RAND_MIX_OFF,
83 FIO_RAND_FILE_OFF,
84 FIO_RAND_BLOCK_OFF,
85 FIO_RAND_FILE_SIZE_OFF,
86 FIO_RAND_TRIM_OFF,
87 FIO_RAND_BUF_OFF,
88 FIO_RAND_SEQ_RAND_READ_OFF,
89 FIO_RAND_SEQ_RAND_WRITE_OFF,
90 FIO_RAND_SEQ_RAND_TRIM_OFF,
91 FIO_RAND_START_DELAY,
92 FIO_RAND_NR_OFFS,
93};
94
95/*
96 * This describes a single thread/process executing a fio job.
97 */
98struct thread_data {
99 struct thread_options o;
100 unsigned long flags;
101 void *eo;
102 char verror[FIO_VERROR_SIZE];
103 pthread_t thread;
104 unsigned int thread_number;
105 unsigned int groupid;
106 struct thread_stat ts;
107
108 int client_type;
109
110 struct io_log *slat_log;
111 struct io_log *clat_log;
112 struct io_log *lat_log;
113 struct io_log *bw_log;
114 struct io_log *iops_log;
115
116 struct tp_data *tp_data;
117
118 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
119 struct timeval bw_sample_time;
120
121 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
122 struct timeval iops_sample_time;
123
124 volatile int update_rusage;
125 struct fio_mutex *rusage_sem;
126 struct rusage ru_start;
127 struct rusage ru_end;
128
129 struct fio_file **files;
130 unsigned char *file_locks;
131 unsigned int files_size;
132 unsigned int files_index;
133 unsigned int nr_open_files;
134 unsigned int nr_done_files;
135 unsigned int nr_normal_files;
136 union {
137 unsigned int next_file;
138 os_random_state_t next_file_state;
139 struct frand_state __next_file_state;
140 };
141 int error;
142 int sig;
143 int done;
144 pid_t pid;
145 char *orig_buffer;
146 size_t orig_buffer_size;
147 volatile int terminate;
148 volatile int runstate;
149 unsigned int last_was_sync;
150 enum fio_ddir last_ddir;
151
152 int mmapfd;
153
154 void *iolog_buf;
155 FILE *iolog_f;
156
157 char *sysfs_root;
158
159 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
160
161 union {
162 os_random_state_t bsrange_state;
163 struct frand_state __bsrange_state;
164 };
165 union {
166 os_random_state_t verify_state;
167 struct frand_state __verify_state;
168 };
169 union {
170 os_random_state_t trim_state;
171 struct frand_state __trim_state;
172 };
173 union {
174 os_random_state_t delay_state;
175 struct frand_state __delay_state;
176 };
177
178 struct frand_state buf_state;
179
180 unsigned int verify_batch;
181 unsigned int trim_batch;
182
183 int shm_id;
184
185 /*
186 * IO engine hooks, contains everything needed to submit an io_u
187 * to any of the available IO engines.
188 */
189 struct ioengine_ops *io_ops;
190
191 /*
192 * Queue depth of io_u's that fio MIGHT do
193 */
194 unsigned int cur_depth;
195
196 /*
197 * io_u's about to be committed
198 */
199 unsigned int io_u_queued;
200
201 /*
202 * io_u's submitted but not completed yet
203 */
204 unsigned int io_u_in_flight;
205
206 /*
207 * List of free and busy io_u's
208 */
209 struct io_u_ring io_u_requeues;
210 struct io_u_queue io_u_freelist;
211 struct io_u_queue io_u_all;
212 pthread_mutex_t io_u_lock;
213 pthread_cond_t free_cond;
214
215 /*
216 * async verify offload
217 */
218 struct flist_head verify_list;
219 pthread_t *verify_threads;
220 unsigned int nr_verify_threads;
221 pthread_cond_t verify_cond;
222 int verify_thread_exit;
223
224 /*
225 * Rate state
226 */
227 uint64_t rate_bps[DDIR_RWDIR_CNT];
228 long rate_pending_usleep[DDIR_RWDIR_CNT];
229 unsigned long rate_bytes[DDIR_RWDIR_CNT];
230 unsigned long rate_blocks[DDIR_RWDIR_CNT];
231 struct timeval lastrate[DDIR_RWDIR_CNT];
232
233 uint64_t total_io_size;
234 uint64_t fill_device_size;
235
236 unsigned long io_issues[DDIR_RWDIR_CNT];
237 uint64_t io_blocks[DDIR_RWDIR_CNT];
238 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
239 uint64_t io_bytes[DDIR_RWDIR_CNT];
240 uint64_t io_skip_bytes;
241 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
242 uint64_t zone_bytes;
243 struct fio_mutex *mutex;
244
245 /*
246 * State for random io, a bitmap of blocks done vs not done
247 */
248 union {
249 os_random_state_t random_state;
250 struct frand_state __random_state;
251 };
252
253 struct timeval start; /* start of this loop */
254 struct timeval epoch; /* time job was started */
255 struct timeval last_issue;
256 struct timeval tv_cache;
257 unsigned int tv_cache_nr;
258 unsigned int tv_cache_mask;
259 unsigned int ramp_time_over;
260
261 /*
262 * Time since last latency_window was started
263 */
264 struct timeval latency_ts;
265 unsigned int latency_qd;
266 unsigned int latency_qd_high;
267 unsigned int latency_qd_low;
268 unsigned int latency_failed;
269 uint64_t latency_ios;
270 int latency_end_run;
271
272 /*
273 * read/write mixed workload state
274 */
275 union {
276 os_random_state_t rwmix_state;
277 struct frand_state __rwmix_state;
278 };
279 unsigned long rwmix_issues;
280 enum fio_ddir rwmix_ddir;
281 unsigned int ddir_seq_nr;
282
283 /*
284 * rand/seq mixed workload state
285 */
286 union {
287 os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
288 struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
289 };
290
291 /*
292 * IO history logs for verification. We use a tree for sorting,
293 * if we are overwriting. Otherwise just use a fifo.
294 */
295 struct rb_root io_hist_tree;
296 struct flist_head io_hist_list;
297 unsigned long io_hist_len;
298
299 /*
300 * For IO replaying
301 */
302 struct flist_head io_log_list;
303
304 /*
305 * For tracking/handling discards
306 */
307 struct flist_head trim_list;
308 unsigned long trim_entries;
309
310 struct flist_head next_rand_list;
311
312 /*
313 * for fileservice, how often to switch to a new file
314 */
315 unsigned int file_service_nr;
316 unsigned int file_service_left;
317 struct fio_file *file_service_file;
318
319 unsigned int sync_file_range_nr;
320
321 /*
322 * For generating file sizes
323 */
324 union {
325 os_random_state_t file_size_state;
326 struct frand_state __file_size_state;
327 };
328
329 /*
330 * Error counts
331 */
332 unsigned int total_err_count;
333 int first_error;
334
335 struct fio_flow *flow;
336
337 /*
338 * Can be overloaded by profiles
339 */
340 struct prof_io_ops prof_io_ops;
341 void *prof_data;
342
343 void *pinned_mem;
344};
345
346/*
347 * when should interactive ETA output be generated
348 */
349enum {
350 FIO_ETA_AUTO,
351 FIO_ETA_ALWAYS,
352 FIO_ETA_NEVER,
353};
354
355#define __td_verror(td, err, msg, func) \
356 do { \
357 unsigned int ____e = (err); \
358 if ((td)->error) \
359 break; \
360 (td)->error = ____e; \
361 if (!(td)->first_error) \
362 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
363 } while (0)
364
365
366#define td_clear_error(td) \
367 (td)->error = 0;
368#define td_verror(td, err, func) \
369 __td_verror((td), (err), strerror((err)), (func))
370#define td_vmsg(td, err, msg, func) \
371 __td_verror((td), (err), (msg), (func))
372
373#define __fio_stringify_1(x) #x
374#define __fio_stringify(x) __fio_stringify_1(x)
375
376extern int exitall_on_terminate;
377extern unsigned int thread_number;
378extern unsigned int stat_number;
379extern int shm_id;
380extern int groupid;
381extern int output_format;
382extern int append_terse_output;
383extern int temp_stall_ts;
384extern uintptr_t page_mask, page_size;
385extern int read_only;
386extern int eta_print;
387extern int eta_new_line;
388extern unsigned long done_secs;
389extern char *job_section;
390extern int fio_gtod_offload;
391extern int fio_gtod_cpu;
392extern enum fio_cs fio_clock_source;
393extern int fio_clock_source_set;
394extern int warnings_fatal;
395extern int terse_version;
396extern int is_backend;
397extern int nr_clients;
398extern int log_syslog;
399extern int status_interval;
400extern const char fio_version_string[];
401
402extern struct thread_data *threads;
403
404static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
405{
406 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
407}
408
409#define REAL_MAX_JOBS 2048
410
411static inline int should_fsync(struct thread_data *td)
412{
413 if (td->last_was_sync)
414 return 0;
415 if (td_write(td) || td_rw(td) || td->o.override_sync)
416 return 1;
417
418 return 0;
419}
420
421/*
422 * Init/option functions
423 */
424extern int __must_check fio_init_options(void);
425extern int __must_check parse_options(int, char **);
426extern int parse_jobs_ini(char *, int, int, int);
427extern int parse_cmd_line(int, char **, int);
428extern int fio_backend(void);
429extern void reset_fio_state(void);
430extern void clear_io_state(struct thread_data *);
431extern int fio_options_parse(struct thread_data *, char **, int, int);
432extern void fio_keywords_init(void);
433extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
434extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
435extern void fio_fill_default_options(struct thread_data *);
436extern int fio_show_option_help(const char *);
437extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
438extern void fio_options_dup_and_init(struct option *);
439extern void fio_options_mem_dupe(struct thread_data *);
440extern void options_mem_dupe(void *data, struct fio_option *options);
441extern void td_fill_rand_seeds(struct thread_data *);
442extern void add_job_opts(const char **, int);
443extern char *num2str(unsigned long, int, int, int, int);
444extern int ioengine_load(struct thread_data *);
445extern int parse_dryrun(void);
446extern int fio_running_or_pending_io_threads(void);
447extern int fio_set_fd_nonblocking(int, const char *);
448
449extern uintptr_t page_mask;
450extern uintptr_t page_size;
451extern int initialize_fio(char *envp[]);
452
453#define FIO_GETOPT_JOB 0x89000000
454#define FIO_GETOPT_IOENGINE 0x98000000
455#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
456
457/*
458 * ETA/status stuff
459 */
460extern void print_thread_status(void);
461extern void print_status_init(int);
462extern char *fio_uint_to_kmg(unsigned int val);
463
464/*
465 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
466 * will never back again. It may cycle between running/verififying/fsyncing.
467 * Once the thread reaches TD_EXITED, it is just waiting for the core to
468 * reap it.
469 */
470enum {
471 TD_NOT_CREATED = 0,
472 TD_CREATED,
473 TD_INITIALIZED,
474 TD_RAMP,
475 TD_SETTING_UP,
476 TD_RUNNING,
477 TD_PRE_READING,
478 TD_VERIFYING,
479 TD_FSYNCING,
480 TD_FINISHING,
481 TD_EXITED,
482 TD_REAPED,
483};
484
485extern void td_set_runstate(struct thread_data *, int);
486extern int td_bump_runstate(struct thread_data *, int);
487extern void td_restore_runstate(struct thread_data *, int);
488
489#define TERMINATE_ALL (-1)
490extern void fio_terminate_threads(int);
491
492/*
493 * Memory helpers
494 */
495extern int __must_check fio_pin_memory(struct thread_data *);
496extern void fio_unpin_memory(struct thread_data *);
497extern int __must_check allocate_io_mem(struct thread_data *);
498extern void free_io_mem(struct thread_data *);
499extern void free_threads_shm(void);
500
501/*
502 * Reset stats after ramp time completes
503 */
504extern void reset_all_stats(struct thread_data *);
505
506/*
507 * blktrace support
508 */
509#ifdef FIO_HAVE_BLKTRACE
510extern int is_blktrace(const char *, int *);
511extern int load_blktrace(struct thread_data *, const char *, int);
512#endif
513
514/*
515 * Latency target helpers
516 */
517extern void lat_target_check(struct thread_data *);
518extern void lat_target_init(struct thread_data *);
519extern void lat_target_reset(struct thread_data *);
520
521#define for_each_td(td, i) \
522 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
523#define for_each_file(td, f, i) \
524 if ((td)->files_index) \
525 for ((i) = 0, (f) = (td)->files[0]; \
526 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
527 (i)++)
528
529#define fio_assert(td, cond) do { \
530 if (!(cond)) { \
531 int *__foo = NULL; \
532 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
533 td_set_runstate((td), TD_EXITED); \
534 (td)->error = EFAULT; \
535 *__foo = 0; \
536 } \
537} while (0)
538
539static inline int fio_fill_issue_time(struct thread_data *td)
540{
541 if (td->o.read_iolog_file ||
542 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
543 return 1;
544
545 return 0;
546}
547
548static inline int __should_check_rate(struct thread_data *td,
549 enum fio_ddir ddir)
550{
551 struct thread_options *o = &td->o;
552
553 /*
554 * If some rate setting was given, we need to check it
555 */
556 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
557 o->rate_iops_min[ddir])
558 return 1;
559
560 return 0;
561}
562
563static inline int should_check_rate(struct thread_data *td,
564 uint64_t *bytes_done)
565{
566 int ret = 0;
567
568 if (bytes_done[DDIR_READ])
569 ret |= __should_check_rate(td, DDIR_READ);
570 if (bytes_done[DDIR_WRITE])
571 ret |= __should_check_rate(td, DDIR_WRITE);
572 if (bytes_done[DDIR_TRIM])
573 ret |= __should_check_rate(td, DDIR_TRIM);
574
575 return ret;
576}
577
578static inline unsigned int td_max_bs(struct thread_data *td)
579{
580 unsigned int max_bs;
581
582 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
583 return max(td->o.max_bs[DDIR_TRIM], max_bs);
584}
585
586static inline unsigned int td_min_bs(struct thread_data *td)
587{
588 unsigned int min_bs;
589
590 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
591 return min(td->o.min_bs[DDIR_TRIM], min_bs);
592}
593
594static inline int is_power_of_2(unsigned long val)
595{
596 return (val != 0 && ((val & (val - 1)) == 0));
597}
598
599/*
600 * We currently only need to do locking if we have verifier threads
601 * accessing our internal structures too
602 */
603static inline void td_io_u_lock(struct thread_data *td)
604{
605 if (td->o.verify_async)
606 pthread_mutex_lock(&td->io_u_lock);
607}
608
609static inline void td_io_u_unlock(struct thread_data *td)
610{
611 if (td->o.verify_async)
612 pthread_mutex_unlock(&td->io_u_lock);
613}
614
615static inline void td_io_u_free_notify(struct thread_data *td)
616{
617 if (td->o.verify_async)
618 pthread_cond_signal(&td->free_cond);
619}
620
621extern const char *fio_get_arch_string(int);
622extern const char *fio_get_os_string(int);
623
624#ifdef FIO_INTERNAL
625#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
626#endif
627
628enum {
629 FIO_OUTPUT_TERSE = 0,
630 FIO_OUTPUT_JSON,
631 FIO_OUTPUT_NORMAL,
632};
633
634enum {
635 FIO_RAND_DIST_RANDOM = 0,
636 FIO_RAND_DIST_ZIPF,
637 FIO_RAND_DIST_PARETO,
638};
639
640enum {
641 FIO_RAND_GEN_TAUSWORTHE = 0,
642 FIO_RAND_GEN_LFSR,
643};
644
645enum {
646 FIO_CPUS_SHARED = 0,
647 FIO_CPUS_SPLIT,
648};
649
650#endif