Makefile: Installing fio2gnuplot & sons
[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 "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};
75
76enum {
77 FIO_RAND_BS_OFF = 0,
78 FIO_RAND_VER_OFF,
79 FIO_RAND_MIX_OFF,
80 FIO_RAND_FILE_OFF,
81 FIO_RAND_BLOCK_OFF,
82 FIO_RAND_FILE_SIZE_OFF,
83 FIO_RAND_TRIM_OFF,
84 FIO_RAND_BUF_OFF,
85 FIO_RAND_SEQ_RAND_READ_OFF,
86 FIO_RAND_SEQ_RAND_WRITE_OFF,
87 FIO_RAND_SEQ_RAND_TRIM_OFF,
88 FIO_RAND_NR_OFFS,
89};
90
91/*
92 * This describes a single thread/process executing a fio job.
93 */
94struct thread_data {
95 struct thread_options o;
96 unsigned long flags;
97 void *eo;
98 char verror[FIO_VERROR_SIZE];
99 pthread_t thread;
100 unsigned int thread_number;
101 unsigned int groupid;
102 struct thread_stat ts;
103
104 int client_type;
105
106 struct io_log *slat_log;
107 struct io_log *clat_log;
108 struct io_log *lat_log;
109 struct io_log *bw_log;
110 struct io_log *iops_log;
111
112 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
113 struct timeval bw_sample_time;
114
115 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
116 struct timeval iops_sample_time;
117
118 volatile int update_rusage;
119 struct fio_mutex *rusage_sem;
120 struct rusage ru_start;
121 struct rusage ru_end;
122
123 struct fio_file **files;
124 unsigned char *file_locks;
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 io_u_ring io_u_requeues;
200 struct io_u_queue io_u_freelist;
201 struct io_u_queue io_u_all;
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 uint64_t 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 uint64_t total_io_size;
224 uint64_t fill_device_size;
225
226 unsigned long io_issues[DDIR_RWDIR_CNT];
227 uint64_t io_blocks[DDIR_RWDIR_CNT];
228 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
229 uint64_t io_bytes[DDIR_RWDIR_CNT];
230 uint64_t io_skip_bytes;
231 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
232 uint64_t 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 * rand/seq mixed workload state
264 */
265 union {
266 os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
267 struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
268 };
269
270 /*
271 * IO history logs for verification. We use a tree for sorting,
272 * if we are overwriting. Otherwise just use a fifo.
273 */
274 struct rb_root io_hist_tree;
275 struct flist_head io_hist_list;
276 unsigned long io_hist_len;
277
278 /*
279 * For IO replaying
280 */
281 struct flist_head io_log_list;
282
283 /*
284 * For tracking/handling discards
285 */
286 struct flist_head trim_list;
287 unsigned long trim_entries;
288
289 struct flist_head next_rand_list;
290
291 /*
292 * for fileservice, how often to switch to a new file
293 */
294 unsigned int file_service_nr;
295 unsigned int file_service_left;
296 struct fio_file *file_service_file;
297
298 unsigned int sync_file_range_nr;
299
300 /*
301 * For generating file sizes
302 */
303 union {
304 os_random_state_t file_size_state;
305 struct frand_state __file_size_state;
306 };
307
308 /*
309 * Error counts
310 */
311 unsigned int total_err_count;
312 int first_error;
313
314 struct fio_flow *flow;
315
316 /*
317 * Can be overloaded by profiles
318 */
319 struct prof_io_ops prof_io_ops;
320 void *prof_data;
321
322 void *pinned_mem;
323};
324
325/*
326 * when should interactive ETA output be generated
327 */
328enum {
329 FIO_ETA_AUTO,
330 FIO_ETA_ALWAYS,
331 FIO_ETA_NEVER,
332};
333
334#define __td_verror(td, err, msg, func) \
335 do { \
336 int e = (err); \
337 if ((td)->error) \
338 break; \
339 (td)->error = e; \
340 if (!(td)->first_error) \
341 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
342 } while (0)
343
344
345#define td_clear_error(td) \
346 (td)->error = 0;
347#define td_verror(td, err, func) \
348 __td_verror((td), (err), strerror((err)), (func))
349#define td_vmsg(td, err, msg, func) \
350 __td_verror((td), (err), (msg), (func))
351
352#define __fio_stringify_1(x) #x
353#define __fio_stringify(x) __fio_stringify_1(x)
354
355extern int exitall_on_terminate;
356extern unsigned int thread_number;
357extern unsigned int stat_number;
358extern int shm_id;
359extern int groupid;
360extern int output_format;
361extern int temp_stall_ts;
362extern uintptr_t page_mask, page_size;
363extern int read_only;
364extern int eta_print;
365extern int eta_new_line;
366extern unsigned long done_secs;
367extern char *job_section;
368extern int fio_gtod_offload;
369extern int fio_gtod_cpu;
370extern enum fio_cs fio_clock_source;
371extern int fio_clock_source_set;
372extern int warnings_fatal;
373extern int terse_version;
374extern int is_backend;
375extern int nr_clients;
376extern int log_syslog;
377extern int status_interval;
378extern const char fio_version_string[];
379
380extern struct thread_data *threads;
381
382static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
383{
384 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
385}
386
387#define REAL_MAX_JOBS 2048
388
389static inline int should_fsync(struct thread_data *td)
390{
391 if (td->last_was_sync)
392 return 0;
393 if (td_write(td) || td_rw(td) || td->o.override_sync)
394 return 1;
395
396 return 0;
397}
398
399/*
400 * Init/option functions
401 */
402extern int __must_check fio_init_options(void);
403extern int __must_check parse_options(int, char **);
404extern int parse_jobs_ini(char *, int, int, int);
405extern int parse_cmd_line(int, char **, int);
406extern int fio_backend(void);
407extern void reset_fio_state(void);
408extern void clear_io_state(struct thread_data *);
409extern int fio_options_parse(struct thread_data *, char **, int);
410extern void fio_keywords_init(void);
411extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
412extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
413extern void fio_fill_default_options(struct thread_data *);
414extern int fio_show_option_help(const char *);
415extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
416extern void fio_options_dup_and_init(struct option *);
417extern void fio_options_mem_dupe(struct thread_data *);
418extern void options_mem_dupe(void *data, struct fio_option *options);
419extern void td_fill_rand_seeds(struct thread_data *);
420extern void add_job_opts(const char **, int);
421extern char *num2str(unsigned long, int, int, int, int);
422extern int ioengine_load(struct thread_data *);
423
424extern uintptr_t page_mask;
425extern uintptr_t page_size;
426extern int initialize_fio(char *envp[]);
427
428#define FIO_GETOPT_JOB 0x89000000
429#define FIO_GETOPT_IOENGINE 0x98000000
430#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
431
432/*
433 * ETA/status stuff
434 */
435extern void print_thread_status(void);
436extern void print_status_init(int);
437extern char *fio_uint_to_kmg(unsigned int val);
438
439/*
440 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
441 * will never back again. It may cycle between running/verififying/fsyncing.
442 * Once the thread reaches TD_EXITED, it is just waiting for the core to
443 * reap it.
444 */
445enum {
446 TD_NOT_CREATED = 0,
447 TD_CREATED,
448 TD_INITIALIZED,
449 TD_RAMP,
450 TD_SETTING_UP,
451 TD_RUNNING,
452 TD_PRE_READING,
453 TD_VERIFYING,
454 TD_FSYNCING,
455 TD_EXITED,
456 TD_REAPED,
457};
458
459extern void td_set_runstate(struct thread_data *, int);
460#define TERMINATE_ALL (-1)
461extern void fio_terminate_threads(int);
462
463/*
464 * Memory helpers
465 */
466extern int __must_check fio_pin_memory(struct thread_data *);
467extern void fio_unpin_memory(struct thread_data *);
468extern int __must_check allocate_io_mem(struct thread_data *);
469extern void free_io_mem(struct thread_data *);
470extern void free_threads_shm(void);
471
472/*
473 * Reset stats after ramp time completes
474 */
475extern void reset_all_stats(struct thread_data *);
476
477/*
478 * blktrace support
479 */
480#ifdef FIO_HAVE_BLKTRACE
481extern int is_blktrace(const char *);
482extern int load_blktrace(struct thread_data *, const char *);
483#endif
484
485#define for_each_td(td, i) \
486 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
487#define for_each_file(td, f, i) \
488 if ((td)->files_index) \
489 for ((i) = 0, (f) = (td)->files[0]; \
490 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
491 (i)++)
492
493#define fio_assert(td, cond) do { \
494 if (!(cond)) { \
495 int *__foo = NULL; \
496 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
497 td_set_runstate((td), TD_EXITED); \
498 (td)->error = EFAULT; \
499 *__foo = 0; \
500 } \
501} while (0)
502
503static inline int fio_fill_issue_time(struct thread_data *td)
504{
505 if (td->o.read_iolog_file ||
506 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
507 return 1;
508
509 return 0;
510}
511
512static inline int __should_check_rate(struct thread_data *td,
513 enum fio_ddir ddir)
514{
515 struct thread_options *o = &td->o;
516
517 /*
518 * If some rate setting was given, we need to check it
519 */
520 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
521 o->rate_iops_min[ddir])
522 return 1;
523
524 return 0;
525}
526
527static inline int should_check_rate(struct thread_data *td,
528 uint64_t *bytes_done)
529{
530 int ret = 0;
531
532 if (bytes_done[DDIR_READ])
533 ret |= __should_check_rate(td, DDIR_READ);
534 if (bytes_done[DDIR_WRITE])
535 ret |= __should_check_rate(td, DDIR_WRITE);
536 if (bytes_done[DDIR_TRIM])
537 ret |= __should_check_rate(td, DDIR_TRIM);
538
539 return ret;
540}
541
542static inline unsigned int td_max_bs(struct thread_data *td)
543{
544 unsigned int max_bs;
545
546 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
547 return max(td->o.max_bs[DDIR_TRIM], max_bs);
548}
549
550static inline unsigned int td_min_bs(struct thread_data *td)
551{
552 unsigned int min_bs;
553
554 min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
555 return min(td->o.min_bs[DDIR_TRIM], min_bs);
556}
557
558static inline int is_power_of_2(unsigned int val)
559{
560 return (val != 0 && ((val & (val - 1)) == 0));
561}
562
563/*
564 * We currently only need to do locking if we have verifier threads
565 * accessing our internal structures too
566 */
567static inline void td_io_u_lock(struct thread_data *td)
568{
569 if (td->o.verify_async)
570 pthread_mutex_lock(&td->io_u_lock);
571}
572
573static inline void td_io_u_unlock(struct thread_data *td)
574{
575 if (td->o.verify_async)
576 pthread_mutex_unlock(&td->io_u_lock);
577}
578
579static inline void td_io_u_free_notify(struct thread_data *td)
580{
581 if (td->o.verify_async)
582 pthread_cond_signal(&td->free_cond);
583}
584
585extern const char *fio_get_arch_string(int);
586extern const char *fio_get_os_string(int);
587
588#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
589
590enum {
591 FIO_OUTPUT_TERSE = 0,
592 FIO_OUTPUT_JSON,
593 FIO_OUTPUT_NORMAL,
594};
595
596enum {
597 FIO_RAND_DIST_RANDOM = 0,
598 FIO_RAND_DIST_ZIPF,
599 FIO_RAND_DIST_PARETO,
600};
601
602enum {
603 FIO_RAND_GEN_TAUSWORTHE = 0,
604 FIO_RAND_GEN_LFSR,
605};
606
607#endif