Merge branch 'master' into gfio
[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
17struct thread_data;
18
19#include "compiler/compiler.h"
20#include "thread_options.h"
21#include "flist.h"
22#include "fifo.h"
23#include "lib/rbtree.h"
24#include "arch/arch.h"
25#include "os/os.h"
26#include "mutex.h"
27#include "log.h"
28#include "debug.h"
29#include "file.h"
30#include "io_ddir.h"
31#include "ioengine.h"
32#include "iolog.h"
33#include "helpers.h"
34#include "options.h"
35#include "profile.h"
36#include "time.h"
37#include "gettime.h"
38#include "lib/getopt.h"
39#include "lib/rand.h"
40#include "client.h"
41#include "server.h"
42#include "stat.h"
43#include "flow.h"
44
45#ifdef FIO_HAVE_GUASI
46#include <guasi.h>
47#endif
48
49#ifdef FIO_HAVE_SOLARISAIO
50#include <sys/asynch.h>
51#endif
52
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
63/*
64 * offset generator types
65 */
66enum {
67 RW_SEQ_SEQ = 0,
68 RW_SEQ_IDENT,
69};
70
71
72enum {
73 TD_F_VER_BACKLOG = 1,
74 TD_F_TRIM_BACKLOG = 2,
75 TD_F_READ_IOLOG = 4,
76 TD_F_REFILL_BUFFERS = 8,
77 TD_F_SCRAMBLE_BUFFERS = 16,
78 TD_F_VER_NONE = 32,
79 TD_F_PROFILE_OPS = 64,
80};
81
82enum {
83 FIO_RAND_BS_OFF = 0,
84 FIO_RAND_VER_OFF,
85 FIO_RAND_MIX_OFF,
86 FIO_RAND_FILE_OFF,
87 FIO_RAND_BLOCK_OFF,
88 FIO_RAND_FILE_SIZE_OFF,
89 FIO_RAND_TRIM_OFF,
90 FIO_RAND_BUF_OFF,
91 FIO_RAND_NR_OFFS,
92};
93
94/*
95 * This describes a single thread/process executing a fio job.
96 */
97struct thread_data {
98 struct thread_options o;
99 unsigned long flags;
100 void *eo;
101 char verror[FIO_VERROR_SIZE];
102 pthread_t thread;
103 unsigned int thread_number;
104 unsigned int groupid;
105 struct thread_stat ts;
106
107 int client_type;
108
109 struct io_log *slat_log;
110 struct io_log *clat_log;
111 struct io_log *lat_log;
112 struct io_log *bw_log;
113 struct io_log *iops_log;
114
115 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
116 struct timeval bw_sample_time;
117
118 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
119 struct timeval iops_sample_time;
120
121 struct rusage ru_start;
122 struct rusage ru_end;
123
124 struct fio_file **files;
125 unsigned int files_size;
126 unsigned int files_index;
127 unsigned int nr_open_files;
128 unsigned int nr_done_files;
129 unsigned int nr_normal_files;
130 union {
131 unsigned int next_file;
132 os_random_state_t next_file_state;
133 struct frand_state __next_file_state;
134 };
135 int error;
136 int sig;
137 int done;
138 pid_t pid;
139 char *orig_buffer;
140 size_t orig_buffer_size;
141 volatile int terminate;
142 volatile int runstate;
143 unsigned int last_was_sync;
144 enum fio_ddir last_ddir;
145
146 int mmapfd;
147
148 void *iolog_buf;
149 FILE *iolog_f;
150
151 char *sysfs_root;
152
153 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
154
155 union {
156 os_random_state_t bsrange_state;
157 struct frand_state __bsrange_state;
158 };
159 union {
160 os_random_state_t verify_state;
161 struct frand_state __verify_state;
162 };
163 union {
164 os_random_state_t trim_state;
165 struct frand_state __trim_state;
166 };
167
168 struct frand_state buf_state;
169
170 unsigned int verify_batch;
171 unsigned int trim_batch;
172
173 int shm_id;
174
175 /*
176 * IO engine hooks, contains everything needed to submit an io_u
177 * to any of the available IO engines.
178 */
179 struct ioengine_ops *io_ops;
180
181 /*
182 * Queue depth of io_u's that fio MIGHT do
183 */
184 unsigned int cur_depth;
185
186 /*
187 * io_u's about to be committed
188 */
189 unsigned int io_u_queued;
190
191 /*
192 * io_u's submitted but not completed yet
193 */
194 unsigned int io_u_in_flight;
195
196 /*
197 * List of free and busy io_u's
198 */
199 struct flist_head io_u_freelist;
200 struct flist_head io_u_busylist;
201 struct flist_head io_u_requeues;
202 pthread_mutex_t io_u_lock;
203 pthread_cond_t free_cond;
204
205 /*
206 * async verify offload
207 */
208 struct flist_head verify_list;
209 pthread_t *verify_threads;
210 unsigned int nr_verify_threads;
211 pthread_cond_t verify_cond;
212 int verify_thread_exit;
213
214 /*
215 * Rate state
216 */
217 unsigned long long rate_bps[DDIR_RWDIR_CNT];
218 long rate_pending_usleep[DDIR_RWDIR_CNT];
219 unsigned long rate_bytes[DDIR_RWDIR_CNT];
220 unsigned long rate_blocks[DDIR_RWDIR_CNT];
221 struct timeval lastrate[DDIR_RWDIR_CNT];
222
223 unsigned long long total_io_size;
224 unsigned long long fill_device_size;
225
226 unsigned long io_issues[DDIR_RWDIR_CNT];
227 unsigned long long io_blocks[DDIR_RWDIR_CNT];
228 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
229 unsigned long long io_bytes[DDIR_RWDIR_CNT];
230 unsigned long long io_skip_bytes;
231 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
232 unsigned long long zone_bytes;
233 struct fio_mutex *mutex;
234
235 /*
236 * State for random io, a bitmap of blocks done vs not done
237 */
238 union {
239 os_random_state_t random_state;
240 struct frand_state __random_state;
241 };
242
243 struct timeval start; /* start of this loop */
244 struct timeval epoch; /* time job was started */
245 struct timeval last_issue;
246 struct timeval tv_cache;
247 unsigned int tv_cache_nr;
248 unsigned int tv_cache_mask;
249 unsigned int ramp_time_over;
250
251 /*
252 * read/write mixed workload state
253 */
254 union {
255 os_random_state_t rwmix_state;
256 struct frand_state __rwmix_state;
257 };
258 unsigned long rwmix_issues;
259 enum fio_ddir rwmix_ddir;
260 unsigned int ddir_seq_nr;
261
262 /*
263 * IO history logs for verification. We use a tree for sorting,
264 * if we are overwriting. Otherwise just use a fifo.
265 */
266 struct rb_root io_hist_tree;
267 struct flist_head io_hist_list;
268 unsigned long io_hist_len;
269
270 /*
271 * For IO replaying
272 */
273 struct flist_head io_log_list;
274
275 /*
276 * For tracking/handling discards
277 */
278 struct flist_head trim_list;
279 unsigned long trim_entries;
280
281 /*
282 * for fileservice, how often to switch to a new file
283 */
284 unsigned int file_service_nr;
285 unsigned int file_service_left;
286 struct fio_file *file_service_file;
287
288 unsigned int sync_file_range_nr;
289
290 /*
291 * For generating file sizes
292 */
293 union {
294 os_random_state_t file_size_state;
295 struct frand_state __file_size_state;
296 };
297
298 /*
299 * Error counts
300 */
301 unsigned int total_err_count;
302 int first_error;
303
304 struct fio_flow *flow;
305
306 /*
307 * Can be overloaded by profiles
308 */
309 struct prof_io_ops prof_io_ops;
310 void *prof_data;
311
312 void *pinned_mem;
313};
314
315/*
316 * when should interactive ETA output be generated
317 */
318enum {
319 FIO_ETA_AUTO,
320 FIO_ETA_ALWAYS,
321 FIO_ETA_NEVER,
322};
323
324#define __td_verror(td, err, msg, func) \
325 do { \
326 if ((td)->error) \
327 break; \
328 int e = (err); \
329 (td)->error = e; \
330 if (!(td)->first_error) \
331 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
332 } while (0)
333
334
335#define td_clear_error(td) \
336 (td)->error = 0;
337#define td_verror(td, err, func) \
338 __td_verror((td), (err), strerror((err)), (func))
339#define td_vmsg(td, err, msg, func) \
340 __td_verror((td), (err), (msg), (func))
341
342#define __fio_stringify_1(x) #x
343#define __fio_stringify(x) __fio_stringify_1(x)
344
345extern int exitall_on_terminate;
346extern unsigned int thread_number;
347extern unsigned int stat_number;
348extern int shm_id;
349extern int groupid;
350extern int output_format;
351extern int temp_stall_ts;
352extern uintptr_t page_mask, page_size;
353extern int read_only;
354extern int eta_print;
355extern unsigned long done_secs;
356extern char *job_section;
357extern int fio_gtod_offload;
358extern int fio_gtod_cpu;
359extern enum fio_cs fio_clock_source;
360extern int fio_clock_source_set;
361extern int warnings_fatal;
362extern int terse_version;
363extern int is_backend;
364extern int nr_clients;
365extern int log_syslog;
366extern const char fio_version_string[];
367extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
368
369extern struct thread_data *threads;
370
371static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
372{
373 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
374}
375
376#define REAL_MAX_JOBS 2048
377
378static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
379{
380 if (err == EILSEQ)
381 return ERROR_TYPE_VERIFY_BIT;
382 if (ddir == DDIR_READ)
383 return ERROR_TYPE_READ_BIT;
384 return ERROR_TYPE_WRITE_BIT;
385}
386
387static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
388static inline int td_non_fatal_error(struct thread_data *td,
389 enum error_type_bit etype, int err)
390{
391 int i;
392 if (!td->o.ignore_error[etype]) {
393 td->o.ignore_error[etype] = __NON_FATAL_ERR;
394 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
395 / sizeof(int);
396 }
397
398 if (!(td->o.continue_on_error & (1 << etype)))
399 return 0;
400 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
401 if (td->o.ignore_error[etype][i] == err)
402 return 1;
403 return 0;
404}
405
406static inline void update_error_count(struct thread_data *td, int err)
407{
408 td->total_err_count++;
409 if (td->total_err_count == 1)
410 td->first_error = err;
411}
412
413static inline int should_fsync(struct thread_data *td)
414{
415 if (td->last_was_sync)
416 return 0;
417 if (td->o.odirect)
418 return 0;
419 if (td_write(td) || td_rw(td) || td->o.override_sync)
420 return 1;
421
422 return 0;
423}
424
425/*
426 * Init/option functions
427 */
428extern int __must_check fio_init_options(void);
429extern int __must_check parse_options(int, char **);
430extern int parse_jobs_ini(char *, int, int, int);
431extern int parse_cmd_line(int, char **, int);
432extern int fio_backend(void);
433extern void reset_fio_state(void);
434extern void clear_io_state(struct thread_data *);
435extern int fio_options_parse(struct thread_data *, char **, int);
436extern void fio_keywords_init(void);
437extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
438extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
439extern void fio_fill_default_options(struct thread_data *);
440extern int fio_show_option_help(const char *);
441extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
442extern void fio_options_dup_and_init(struct option *);
443extern void fio_options_mem_dupe(struct thread_data *);
444extern void options_mem_dupe(void *data, struct fio_option *options);
445extern void td_fill_rand_seeds(struct thread_data *);
446extern void add_job_opts(const char **, int);
447extern char *num2str(unsigned long, int, int, int);
448extern int ioengine_load(struct thread_data *);
449
450extern unsigned long page_mask;
451extern unsigned long page_size;
452extern int initialize_fio(char *envp[]);
453
454#define FIO_GETOPT_JOB 0x89000000
455#define FIO_GETOPT_IOENGINE 0x98000000
456#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
457
458/*
459 * ETA/status stuff
460 */
461extern void print_thread_status(void);
462extern void print_status_init(int);
463extern char *fio_uint_to_kmg(unsigned int val);
464
465/*
466 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
467 * will never back again. It may cycle between running/verififying/fsyncing.
468 * Once the thread reaches TD_EXITED, it is just waiting for the core to
469 * reap it.
470 */
471enum {
472 TD_NOT_CREATED = 0,
473 TD_CREATED,
474 TD_INITIALIZED,
475 TD_RAMP,
476 TD_SETTING_UP,
477 TD_RUNNING,
478 TD_PRE_READING,
479 TD_VERIFYING,
480 TD_FSYNCING,
481 TD_EXITED,
482 TD_REAPED,
483};
484
485extern void td_set_runstate(struct thread_data *, int);
486#define TERMINATE_ALL (-1)
487extern void fio_terminate_threads(int);
488
489/*
490 * Memory helpers
491 */
492extern int __must_check fio_pin_memory(struct thread_data *);
493extern void fio_unpin_memory(struct thread_data *);
494extern int __must_check allocate_io_mem(struct thread_data *);
495extern void free_io_mem(struct thread_data *);
496extern void free_threads_shm(void);
497
498/*
499 * Reset stats after ramp time completes
500 */
501extern void reset_all_stats(struct thread_data *);
502
503/*
504 * blktrace support
505 */
506#ifdef FIO_HAVE_BLKTRACE
507extern int is_blktrace(const char *);
508extern int load_blktrace(struct thread_data *, const char *);
509#endif
510
511/*
512 * Mark unused variables passed to ops functions as unused, to silence gcc
513 */
514#define fio_unused __attribute((__unused__))
515#define fio_init __attribute__((constructor))
516#define fio_exit __attribute__((destructor))
517
518#define for_each_td(td, i) \
519 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
520#define for_each_file(td, f, i) \
521 if ((td)->files_index) \
522 for ((i) = 0, (f) = (td)->files[0]; \
523 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
524 (i)++)
525
526#define fio_assert(td, cond) do { \
527 if (!(cond)) { \
528 int *__foo = NULL; \
529 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
530 td_set_runstate((td), TD_EXITED); \
531 (td)->error = EFAULT; \
532 *__foo = 0; \
533 } \
534} while (0)
535
536static inline int fio_fill_issue_time(struct thread_data *td)
537{
538 if (td->o.read_iolog_file ||
539 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
540 return 1;
541
542 return 0;
543}
544
545static inline int __should_check_rate(struct thread_data *td,
546 enum fio_ddir ddir)
547{
548 struct thread_options *o = &td->o;
549
550 /*
551 * If some rate setting was given, we need to check it
552 */
553 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
554 o->rate_iops_min[ddir])
555 return 1;
556
557 return 0;
558}
559
560static inline int should_check_rate(struct thread_data *td,
561 unsigned long *bytes_done)
562{
563 int ret = 0;
564
565 if (bytes_done[DDIR_READ])
566 ret |= __should_check_rate(td, DDIR_READ);
567 if (bytes_done[DDIR_WRITE])
568 ret |= __should_check_rate(td, DDIR_WRITE);
569 if (bytes_done[DDIR_TRIM])
570 ret |= __should_check_rate(td, DDIR_TRIM);
571
572 return ret;
573}
574
575static inline int is_power_of_2(unsigned int val)
576{
577 return (val != 0 && ((val & (val - 1)) == 0));
578}
579
580/*
581 * We currently only need to do locking if we have verifier threads
582 * accessing our internal structures too
583 */
584static inline void td_io_u_lock(struct thread_data *td)
585{
586 if (td->o.verify_async)
587 pthread_mutex_lock(&td->io_u_lock);
588}
589
590static inline void td_io_u_unlock(struct thread_data *td)
591{
592 if (td->o.verify_async)
593 pthread_mutex_unlock(&td->io_u_lock);
594}
595
596static inline void td_io_u_free_notify(struct thread_data *td)
597{
598 if (td->o.verify_async)
599 pthread_cond_signal(&td->free_cond);
600}
601
602extern const char *fio_get_arch_string(int);
603extern const char *fio_get_os_string(int);
604
605#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
606
607enum {
608 FIO_OUTPUT_TERSE = 0,
609 FIO_OUTPUT_JSON,
610 FIO_OUTPUT_NORMAL,
611};
612
613enum {
614 FIO_RAND_DIST_RANDOM = 0,
615 FIO_RAND_DIST_ZIPF,
616 FIO_RAND_DIST_PARETO,
617};
618
619enum {
620 FIO_RAND_GEN_TAUSWORTHE = 0,
621 FIO_RAND_GEN_LFSR,
622};
623
624#endif