server: properly reset state_number in fio_reset_state()
[fio.git] / fio.h
CommitLineData
ebac4655
JA
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>
3c39a379
JA
9#include <errno.h>
10#include <stdlib.h>
11#include <stdio.h>
6d6f031f 12#include <unistd.h>
34cfcdaf 13#include <string.h>
cd14cc10 14#include <inttypes.h>
7101d9c2 15#include <assert.h>
ebac4655 16
ecc314ba
BC
17struct thread_data;
18
317b95d0 19#include "compiler/compiler.h"
ec41265e 20#include "thread_options.h"
01743ee1 21#include "flist.h"
e2887563 22#include "fifo.h"
05775438 23#include "lib/rbtree.h"
317b95d0
JA
24#include "arch/arch.h"
25#include "os/os.h"
07739b57 26#include "mutex.h"
a3d741fa
JA
27#include "log.h"
28#include "debug.h"
d6aed795
JA
29#include "file.h"
30#include "io_ddir.h"
dcefb588 31#include "ioengine.h"
5995a6a4 32#include "iolog.h"
c5c8bd5c 33#include "helpers.h"
07b3232d 34#include "options.h"
7eb36574 35#include "profile.h"
c223da83 36#include "time.h"
a7346816 37#include "gettime.h"
bf2e821a 38#include "lib/getopt.h"
2615cc4b 39#include "lib/rand.h"
dd366728 40#include "client.h"
37db14fe 41#include "server.h"
a64e88da 42#include "stat.h"
9e684a49 43#include "flow.h"
ebac4655 44
609342ff
DL
45#ifdef FIO_HAVE_GUASI
46#include <guasi.h>
47#endif
48
417f0068
JA
49#ifdef FIO_HAVE_SOLARISAIO
50#include <sys/asynch.h>
51#endif
52
d0b937ed
YR
53#ifdef FIO_HAVE_LIBNUMA
54#include <linux/mempolicy.h>
55#include <numa.h>
56
57/*
58 * "local" is pseudo-policy
59 */
60#define MPOL_LOCAL MPOL_MAX
61#endif
62
38dad62d
JA
63/*
64 * offset generator types
65 */
66enum {
67 RW_SEQ_SEQ = 0,
68 RW_SEQ_IDENT,
69};
70
2dc1bbeb
JA
71/*
72 * This describes a single thread/process executing a fio job.
73 */
74struct thread_data {
75 struct thread_options o;
de890a1e 76 void *eo;
e4e33258 77 char verror[FIO_VERROR_SIZE];
2dc1bbeb 78 pthread_t thread;
2f122b13
JA
79 unsigned int thread_number;
80 unsigned int groupid;
2dc1bbeb 81 struct thread_stat ts;
7b9f733a 82
46bcd498
JA
83 int client_type;
84
7b9f733a
JA
85 struct io_log *slat_log;
86 struct io_log *clat_log;
87 struct io_log *lat_log;
88 struct io_log *bw_log;
c8eeb9df 89 struct io_log *iops_log;
7b9f733a 90
6eaf09d6 91 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
c8eeb9df
JA
92 struct timeval bw_sample_time;
93
6eaf09d6 94 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
c8eeb9df 95 struct timeval iops_sample_time;
f0505a14 96
c8aaba19
JA
97 struct rusage ru_start;
98 struct rusage ru_end;
99
126d65c6 100 struct fio_file **files;
dd87b2c9 101 unsigned int files_size;
2dc1bbeb
JA
102 unsigned int files_index;
103 unsigned int nr_open_files;
1020a139 104 unsigned int nr_done_files;
2dc1bbeb
JA
105 unsigned int nr_normal_files;
106 union {
107 unsigned int next_file;
108 os_random_state_t next_file_state;
4c07ad86 109 struct frand_state __next_file_state;
2dc1bbeb
JA
110 };
111 int error;
a5e371a6 112 int sig;
20e354ef 113 int done;
2dc1bbeb
JA
114 pid_t pid;
115 char *orig_buffer;
116 size_t orig_buffer_size;
117 volatile int terminate;
118 volatile int runstate;
2dc1bbeb 119 unsigned int last_was_sync;
9e144189 120 enum fio_ddir last_ddir;
2dc1bbeb 121
2dc1bbeb
JA
122 int mmapfd;
123
843a7413
JA
124 void *iolog_buf;
125 FILE *iolog_f;
ebac4655 126
da86774e 127 char *sysfs_root;
da86774e 128
3545a109 129 unsigned long rand_seeds[8];
5bfc35d7 130
4c07ad86
JA
131 union {
132 os_random_state_t bsrange_state;
133 struct frand_state __bsrange_state;
134 };
135 union {
136 os_random_state_t verify_state;
137 struct frand_state __verify_state;
138 };
139 union {
140 os_random_state_t trim_state;
141 struct frand_state __trim_state;
142 };
ebac4655 143
3545a109
JA
144 struct frand_state buf_state;
145
9e144189 146 unsigned int verify_batch;
0d29de83 147 unsigned int trim_batch;
9e144189 148
ebac4655
JA
149 int shm_id;
150
e9c047a0
JA
151 /*
152 * IO engine hooks, contains everything needed to submit an io_u
153 * to any of the available IO engines.
154 */
2866c82d 155 struct ioengine_ops *io_ops;
ebac4655 156
e9c047a0 157 /*
422f9e4b 158 * Queue depth of io_u's that fio MIGHT do
e9c047a0 159 */
ebac4655 160 unsigned int cur_depth;
422f9e4b
RM
161
162 /*
163 * io_u's about to be committed
164 */
d8005759 165 unsigned int io_u_queued;
422f9e4b
RM
166
167 /*
168 * io_u's submitted but not completed yet
169 */
170 unsigned int io_u_in_flight;
171
172 /*
173 * List of free and busy io_u's
174 */
01743ee1
JA
175 struct flist_head io_u_freelist;
176 struct flist_head io_u_busylist;
177 struct flist_head io_u_requeues;
e8462bd8
JA
178 pthread_mutex_t io_u_lock;
179 pthread_cond_t free_cond;
180
181 /*
182 * async verify offload
183 */
184 struct flist_head verify_list;
185 pthread_t *verify_threads;
186 unsigned int nr_verify_threads;
187 pthread_cond_t verify_cond;
188 int verify_thread_exit;
ebac4655 189
e9c047a0
JA
190 /*
191 * Rate state
192 */
6eaf09d6
SL
193 unsigned long long rate_bps[DDIR_RWDIR_CNT];
194 long rate_pending_usleep[DDIR_RWDIR_CNT];
195 unsigned long rate_bytes[DDIR_RWDIR_CNT];
196 unsigned long rate_blocks[DDIR_RWDIR_CNT];
197 struct timeval lastrate[DDIR_RWDIR_CNT];
ebac4655 198
ebac4655 199 unsigned long long total_io_size;
2e3bd4c2 200 unsigned long long fill_device_size;
ebac4655 201
6eaf09d6
SL
202 unsigned long io_issues[DDIR_RWDIR_CNT];
203 unsigned long long io_blocks[DDIR_RWDIR_CNT];
204 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
205 unsigned long long io_bytes[DDIR_RWDIR_CNT];
48f5abd3 206 unsigned long long io_skip_bytes;
6eaf09d6 207 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
079ad09b 208 unsigned long long zone_bytes;
cdd18ad8 209 struct fio_mutex *mutex;
ebac4655 210
e9c047a0
JA
211 /*
212 * State for random io, a bitmap of blocks done vs not done
213 */
4c07ad86
JA
214 union {
215 os_random_state_t random_state;
216 struct frand_state __random_state;
217 };
ebac4655 218
ebac4655
JA
219 struct timeval start; /* start of this loop */
220 struct timeval epoch; /* time job was started */
a61eddec 221 struct timeval last_issue;
993bf48b
JA
222 struct timeval tv_cache;
223 unsigned int tv_cache_nr;
224 unsigned int tv_cache_mask;
721938ae 225 unsigned int ramp_time_over;
ebac4655 226
e9c047a0
JA
227 /*
228 * read/write mixed workload state
229 */
4c07ad86
JA
230 union {
231 os_random_state_t rwmix_state;
232 struct frand_state __rwmix_state;
233 };
e4928662 234 unsigned long rwmix_issues;
e9c047a0 235 enum fio_ddir rwmix_ddir;
5736c10d 236 unsigned int ddir_seq_nr;
a6ccc7be 237
e9c047a0 238 /*
8de8f047
JA
239 * IO history logs for verification. We use a tree for sorting,
240 * if we are overwriting. Otherwise just use a fifo.
e9c047a0 241 */
4b87898e 242 struct rb_root io_hist_tree;
01743ee1 243 struct flist_head io_hist_list;
9e144189 244 unsigned long io_hist_len;
8de8f047
JA
245
246 /*
247 * For IO replaying
248 */
01743ee1 249 struct flist_head io_log_list;
433afcb4 250
0d29de83
JA
251 /*
252 * For tracking/handling discards
253 */
254 struct flist_head trim_list;
255 unsigned long trim_entries;
256
1907dbc6
JA
257 /*
258 * for fileservice, how often to switch to a new file
259 */
260 unsigned int file_service_nr;
261 unsigned int file_service_left;
262 struct fio_file *file_service_file;
9c60ce64 263
44f29692
JA
264 unsigned int sync_file_range_nr;
265
9c60ce64
JA
266 /*
267 * For generating file sizes
268 */
4c07ad86
JA
269 union {
270 os_random_state_t file_size_state;
271 struct frand_state __file_size_state;
272 };
f2bba182
RR
273
274 /*
275 * Error counts
276 */
277 unsigned int total_err_count;
278 int first_error;
15dc1934 279
9e684a49
DE
280 struct fio_flow *flow;
281
15dc1934
JA
282 /*
283 * Can be overloaded by profiles
284 */
7eb36574 285 struct prof_io_ops prof_io_ops;
58c55ba0 286 void *prof_data;
1b79a070
JA
287
288 void *pinned_mem;
ebac4655
JA
289};
290
e592a06b
AC
291/*
292 * when should interactive ETA output be generated
293 */
294enum {
295 FIO_ETA_AUTO,
296 FIO_ETA_ALWAYS,
297 FIO_ETA_NEVER,
298};
299
e1161c32 300#define __td_verror(td, err, msg, func) \
ebac4655 301 do { \
19abcd3d
JA
302 if ((td)->error) \
303 break; \
ebac4655
JA
304 int e = (err); \
305 (td)->error = e; \
f2bba182
RR
306 if (!(td)->first_error) \
307 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
ebac4655
JA
308 } while (0)
309
b990b5c0 310
f2bba182
RR
311#define td_clear_error(td) \
312 (td)->error = 0;
e1161c32
JA
313#define td_verror(td, err, func) \
314 __td_verror((td), (err), strerror((err)), (func))
315#define td_vmsg(td, err, msg, func) \
316 __td_verror((td), (err), (msg), (func))
b990b5c0 317
81179eec
JA
318#define __fio_stringify_1(x) #x
319#define __fio_stringify(x) __fio_stringify_1(x)
320
ebac4655 321extern int exitall_on_terminate;
11e950bd 322extern unsigned int thread_number;
108fea77 323extern unsigned int stat_number;
ebac4655
JA
324extern int shm_id;
325extern int groupid;
f3afa57e 326extern int output_format;
53cdc686 327extern int temp_stall_ts;
7078106d 328extern uintptr_t page_mask, page_size;
4241ea8f 329extern int read_only;
e592a06b 330extern int eta_print;
d3eeeabc 331extern unsigned long done_secs;
01f06b63 332extern char *job_section;
be4ecfdf
JA
333extern int fio_gtod_offload;
334extern int fio_gtod_cpu;
c223da83 335extern enum fio_cs fio_clock_source;
a9523c6f 336extern int warnings_fatal;
f57a9c59 337extern int terse_version;
5c341e9a 338extern int is_backend;
a37f69b7 339extern int nr_clients;
e46d8091 340extern int log_syslog;
5e726d0a 341extern const char fio_version_string[];
802ad4a8 342extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
ebac4655
JA
343
344extern struct thread_data *threads;
345
7101d9c2
JA
346static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
347{
348 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
349}
350
0ce8b119 351#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
aec2de20
JA
352#define TO_MAP_BLOCK(f, b) (b)
353#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
354#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
ebac4655 355
fca70358 356#define REAL_MAX_JOBS 2048
ebac4655 357
4373e97a 358static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
06842027
SL
359{
360 if (err == EILSEQ)
8b28bd41 361 return ERROR_TYPE_VERIFY_BIT;
06842027 362 if (ddir == DDIR_READ)
8b28bd41
DM
363 return ERROR_TYPE_READ_BIT;
364 return ERROR_TYPE_WRITE_BIT;
365}
366
367static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
368static inline int td_non_fatal_error(struct thread_data *td,
369 enum error_type_bit etype, int err)
370{
371 int i;
372 if (!td->o.ignore_error[etype]) {
373 td->o.ignore_error[etype] = __NON_FATAL_ERR;
374 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
375 / sizeof(int);
376 }
377
378 if (!(td->o.continue_on_error & (1 << etype)))
379 return 0;
380 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
381 if (td->o.ignore_error[etype][i] == err)
382 return 1;
383 return 0;
06842027
SL
384}
385
f2bba182
RR
386static inline void update_error_count(struct thread_data *td, int err)
387{
388 td->total_err_count++;
389 if (td->total_err_count == 1)
390 td->first_error = err;
391}
392
87dc1ab1
JA
393static inline int should_fsync(struct thread_data *td)
394{
395 if (td->last_was_sync)
396 return 0;
2dc1bbeb 397 if (td->o.odirect)
87dc1ab1 398 return 0;
2dc1bbeb 399 if (td_write(td) || td_rw(td) || td->o.override_sync)
87dc1ab1
JA
400 return 1;
401
402 return 0;
403}
404
8914a9d8 405/*
214e1eca 406 * Init/option functions
8914a9d8 407 */
0420ba6a 408extern int __must_check fio_init_options(void);
72cb971b 409extern int __must_check parse_options(int, char **);
46bcd498
JA
410extern int parse_jobs_ini(char *, int, int, int);
411extern int parse_cmd_line(int, char **, int);
2e1df07d 412extern int fio_backend(void);
50d16976 413extern void reset_fio_state(void);
2e1df07d 414extern void clear_io_state(struct thread_data *);
3b8b7135 415extern int fio_options_parse(struct thread_data *, char **, int);
74929ac2 416extern void fio_keywords_init(void);
214e1eca 417extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
de890a1e 418extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
214e1eca
JA
419extern void fio_fill_default_options(struct thread_data *);
420extern int fio_show_option_help(const char *);
de890a1e 421extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
214e1eca 422extern void fio_options_dup_and_init(struct option *);
de890a1e
SL
423extern void fio_options_mem_dupe(struct thread_data *);
424extern void options_mem_dupe(void *data, struct fio_option *options);
5bfc35d7 425extern void td_fill_rand_seeds(struct thread_data *);
46bcd498 426extern void add_job_opts(const char **, int);
1ec3d69b 427extern char *num2str(unsigned long, int, int, int);
de890a1e 428extern int ioengine_load(struct thread_data *);
1ec3d69b 429
e61ca217
JA
430extern unsigned long page_mask;
431extern unsigned long page_size;
432extern int initialize_fio(char *envp[]);
433
de890a1e
SL
434#define FIO_GETOPT_JOB 0x89000000
435#define FIO_GETOPT_IOENGINE 0x98000000
07b3232d 436#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
8914a9d8 437
263e529f
JA
438/*
439 * ETA/status stuff
440 */
441extern void print_thread_status(void);
442extern void print_status_init(int);
99d633af 443extern char *fio_uint_to_kmg(unsigned int val);
263e529f
JA
444
445/*
446 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
447 * will never back again. It may cycle between running/verififying/fsyncing.
448 * Once the thread reaches TD_EXITED, it is just waiting for the core to
449 * reap it.
450 */
451enum {
452 TD_NOT_CREATED = 0,
453 TD_CREATED,
454 TD_INITIALIZED,
9c6f6316 455 TD_SETTING_UP,
b29ee5b3 456 TD_RAMP,
263e529f 457 TD_RUNNING,
b0f65863 458 TD_PRE_READING,
263e529f
JA
459 TD_VERIFYING,
460 TD_FSYNCING,
461 TD_EXITED,
462 TD_REAPED,
463};
464
b29ee5b3 465extern void td_set_runstate(struct thread_data *, int);
cc0df00a
JA
466#define TERMINATE_ALL (-1)
467extern void fio_terminate_threads(int);
b29ee5b3 468
2f9ade3c
JA
469/*
470 * Memory helpers
471 */
1b79a070
JA
472extern int __must_check fio_pin_memory(struct thread_data *);
473extern void fio_unpin_memory(struct thread_data *);
b2fdda43 474extern int __must_check allocate_io_mem(struct thread_data *);
2f9ade3c 475extern void free_io_mem(struct thread_data *);
2bb3f0a7 476extern void free_threads_shm(void);
2f9ade3c 477
b29ee5b3
JA
478/*
479 * Reset stats after ramp time completes
480 */
481extern void reset_all_stats(struct thread_data *);
482
fb7b71a3
JA
483/*
484 * blktrace support
485 */
5e62c22a 486#ifdef FIO_HAVE_BLKTRACE
fb7b71a3
JA
487extern int is_blktrace(const char *);
488extern int load_blktrace(struct thread_data *, const char *);
5e62c22a 489#endif
fb7b71a3 490
2866c82d
JA
491/*
492 * Mark unused variables passed to ops functions as unused, to silence gcc
493 */
494#define fio_unused __attribute((__unused__))
5f350952
JA
495#define fio_init __attribute__((constructor))
496#define fio_exit __attribute__((destructor))
2866c82d 497
34572e28
JA
498#define for_each_td(td, i) \
499 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
53cdc686 500#define for_each_file(td, f, i) \
691c8fb0
JA
501 if ((td)->files_index) \
502 for ((i) = 0, (f) = (td)->files[0]; \
503 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
504 (i)++)
53cdc686 505
0032bf9f
JA
506#define fio_assert(td, cond) do { \
507 if (!(cond)) { \
340fd243 508 int *__foo = NULL; \
0032bf9f 509 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
ac18ea38 510 td_set_runstate((td), TD_EXITED); \
437c9b71 511 (td)->error = EFAULT; \
340fd243 512 *__foo = 0; \
0032bf9f
JA
513 } \
514} while (0)
515
12d9d841
JA
516static inline int fio_fill_issue_time(struct thread_data *td)
517{
518 if (td->o.read_iolog_file ||
519 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
520 return 1;
521
522 return 0;
523}
524
581e7141
JA
525static inline int __should_check_rate(struct thread_data *td,
526 enum fio_ddir ddir)
527{
528 struct thread_options *o = &td->o;
529
530 /*
531 * If some rate setting was given, we need to check it
532 */
533 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
534 o->rate_iops_min[ddir])
535 return 1;
536
537 return 0;
538}
539
540static inline int should_check_rate(struct thread_data *td,
541 unsigned long *bytes_done)
542{
543 int ret = 0;
544
6eaf09d6
SL
545 if (bytes_done[DDIR_READ])
546 ret |= __should_check_rate(td, DDIR_READ);
547 if (bytes_done[DDIR_WRITE])
548 ret |= __should_check_rate(td, DDIR_WRITE);
549 if (bytes_done[DDIR_TRIM])
550 ret |= __should_check_rate(td, DDIR_TRIM);
581e7141
JA
551
552 return ret;
553}
554
d529ee19
JA
555static inline int is_power_of_2(unsigned int val)
556{
557 return (val != 0 && ((val & (val - 1)) == 0));
558}
559
e8462bd8
JA
560/*
561 * We currently only need to do locking if we have verifier threads
562 * accessing our internal structures too
563 */
564static inline void td_io_u_lock(struct thread_data *td)
565{
566 if (td->o.verify_async)
567 pthread_mutex_lock(&td->io_u_lock);
568}
569
570static inline void td_io_u_unlock(struct thread_data *td)
571{
572 if (td->o.verify_async)
573 pthread_mutex_unlock(&td->io_u_lock);
574}
575
576static inline void td_io_u_free_notify(struct thread_data *td)
577{
578 if (td->o.verify_async)
579 pthread_cond_signal(&td->free_cond);
580}
581
cca84643
JA
582extern const char *fio_get_arch_string(int);
583extern const char *fio_get_os_string(int);
584
05775438
JA
585#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
586
f3afa57e
JA
587enum {
588 FIO_OUTPUT_TERSE = 0,
589 FIO_OUTPUT_JSON,
590 FIO_OUTPUT_NORMAL,
591};
592
e25839d4
JA
593enum {
594 FIO_RAND_DIST_RANDOM = 0,
595 FIO_RAND_DIST_ZIPF,
925fee33 596 FIO_RAND_DIST_PARETO,
e25839d4
JA
597};
598
ebac4655 599#endif