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/*
54 * offset generator types
55 */
56enum {
57 RW_SEQ_SEQ = 0,
58 RW_SEQ_IDENT,
59};
60
61/*
62 * This describes a single thread/process executing a fio job.
63 */
64struct thread_data {
65 struct thread_options o;
66 void *eo;
67 char verror[FIO_VERROR_SIZE];
68 pthread_t thread;
69 unsigned int thread_number;
70 unsigned int groupid;
71 struct thread_stat ts;
72
73 int client_type;
74
75 struct io_log *slat_log;
76 struct io_log *clat_log;
77 struct io_log *lat_log;
78 struct io_log *bw_log;
79 struct io_log *iops_log;
80
81 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
82 struct timeval bw_sample_time;
83
84 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
85 struct timeval iops_sample_time;
86
87 struct rusage ru_start;
88 struct rusage ru_end;
89
90 struct fio_file **files;
91 unsigned int files_size;
92 unsigned int files_index;
93 unsigned int nr_open_files;
94 unsigned int nr_done_files;
95 unsigned int nr_normal_files;
96 union {
97 unsigned int next_file;
98 os_random_state_t next_file_state;
99 struct frand_state __next_file_state;
100 };
101 int error;
102 int sig;
103 int done;
104 pid_t pid;
105 char *orig_buffer;
106 size_t orig_buffer_size;
107 volatile int terminate;
108 volatile int runstate;
109 unsigned int last_was_sync;
110 enum fio_ddir last_ddir;
111
112 int mmapfd;
113
114 void *iolog_buf;
115 FILE *iolog_f;
116
117 char *sysfs_root;
118
119 unsigned long rand_seeds[8];
120
121 union {
122 os_random_state_t bsrange_state;
123 struct frand_state __bsrange_state;
124 };
125 union {
126 os_random_state_t verify_state;
127 struct frand_state __verify_state;
128 };
129 union {
130 os_random_state_t trim_state;
131 struct frand_state __trim_state;
132 };
133
134 struct frand_state buf_state;
135
136 unsigned int verify_batch;
137 unsigned int trim_batch;
138
139 int shm_id;
140
141 /*
142 * IO engine hooks, contains everything needed to submit an io_u
143 * to any of the available IO engines.
144 */
145 struct ioengine_ops *io_ops;
146
147 /*
148 * Queue depth of io_u's that fio MIGHT do
149 */
150 unsigned int cur_depth;
151
152 /*
153 * io_u's about to be committed
154 */
155 unsigned int io_u_queued;
156
157 /*
158 * io_u's submitted but not completed yet
159 */
160 unsigned int io_u_in_flight;
161
162 /*
163 * List of free and busy io_u's
164 */
165 struct flist_head io_u_freelist;
166 struct flist_head io_u_busylist;
167 struct flist_head io_u_requeues;
168 pthread_mutex_t io_u_lock;
169 pthread_cond_t free_cond;
170
171 /*
172 * async verify offload
173 */
174 struct flist_head verify_list;
175 pthread_t *verify_threads;
176 unsigned int nr_verify_threads;
177 pthread_cond_t verify_cond;
178 int verify_thread_exit;
179
180 /*
181 * Rate state
182 */
183 unsigned long long rate_bps[DDIR_RWDIR_CNT];
184 long rate_pending_usleep[DDIR_RWDIR_CNT];
185 unsigned long rate_bytes[DDIR_RWDIR_CNT];
186 unsigned long rate_blocks[DDIR_RWDIR_CNT];
187 struct timeval lastrate[DDIR_RWDIR_CNT];
188
189 unsigned long long total_io_size;
190 unsigned long long fill_device_size;
191
192 unsigned long io_issues[DDIR_RWDIR_CNT];
193 unsigned long long io_blocks[DDIR_RWDIR_CNT];
194 unsigned long long this_io_blocks[DDIR_RWDIR_CNT];
195 unsigned long long io_bytes[DDIR_RWDIR_CNT];
196 unsigned long long io_skip_bytes;
197 unsigned long long this_io_bytes[DDIR_RWDIR_CNT];
198 unsigned long long zone_bytes;
199 struct fio_mutex *mutex;
200
201 /*
202 * State for random io, a bitmap of blocks done vs not done
203 */
204 union {
205 os_random_state_t random_state;
206 struct frand_state __random_state;
207 };
208
209 struct timeval start; /* start of this loop */
210 struct timeval epoch; /* time job was started */
211 struct timeval last_issue;
212 struct timeval tv_cache;
213 unsigned int tv_cache_nr;
214 unsigned int tv_cache_mask;
215 unsigned int ramp_time_over;
216
217 /*
218 * read/write mixed workload state
219 */
220 union {
221 os_random_state_t rwmix_state;
222 struct frand_state __rwmix_state;
223 };
224 unsigned long rwmix_issues;
225 enum fio_ddir rwmix_ddir;
226 unsigned int ddir_seq_nr;
227
228 /*
229 * IO history logs for verification. We use a tree for sorting,
230 * if we are overwriting. Otherwise just use a fifo.
231 */
232 struct rb_root io_hist_tree;
233 struct flist_head io_hist_list;
234 unsigned long io_hist_len;
235
236 /*
237 * For IO replaying
238 */
239 struct flist_head io_log_list;
240
241 /*
242 * For tracking/handling discards
243 */
244 struct flist_head trim_list;
245 unsigned long trim_entries;
246
247 /*
248 * for fileservice, how often to switch to a new file
249 */
250 unsigned int file_service_nr;
251 unsigned int file_service_left;
252 struct fio_file *file_service_file;
253
254 unsigned int sync_file_range_nr;
255
256 /*
257 * For generating file sizes
258 */
259 union {
260 os_random_state_t file_size_state;
261 struct frand_state __file_size_state;
262 };
263
264 /*
265 * Error counts
266 */
267 unsigned int total_err_count;
268 int first_error;
269
270 struct fio_flow *flow;
271
272 /*
273 * Can be overloaded by profiles
274 */
275 struct prof_io_ops prof_io_ops;
276 void *prof_data;
277
278 void *pinned_mem;
279};
280
281/*
282 * when should interactive ETA output be generated
283 */
284enum {
285 FIO_ETA_AUTO,
286 FIO_ETA_ALWAYS,
287 FIO_ETA_NEVER,
288};
289
290#define __td_verror(td, err, msg, func) \
291 do { \
292 if ((td)->error) \
293 break; \
294 int e = (err); \
295 (td)->error = e; \
296 if (!(td)->first_error) \
297 snprintf(td->verror, sizeof(td->verror) - 1, "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
298 } while (0)
299
300
301#define td_clear_error(td) \
302 (td)->error = 0;
303#define td_verror(td, err, func) \
304 __td_verror((td), (err), strerror((err)), (func))
305#define td_vmsg(td, err, msg, func) \
306 __td_verror((td), (err), (msg), (func))
307
308#define __fio_stringify_1(x) #x
309#define __fio_stringify(x) __fio_stringify_1(x)
310
311extern int exitall_on_terminate;
312extern unsigned int thread_number;
313extern int shm_id;
314extern int groupid;
315extern int output_format;
316extern int temp_stall_ts;
317extern uintptr_t page_mask, page_size;
318extern int read_only;
319extern int eta_print;
320extern unsigned long done_secs;
321extern char *job_section;
322extern int fio_gtod_offload;
323extern int fio_gtod_cpu;
324extern enum fio_cs fio_clock_source;
325extern int warnings_fatal;
326extern int terse_version;
327extern int is_backend;
328extern int nr_clients;
329extern int log_syslog;
330extern const char fio_version_string[];
331extern const fio_fp64_t def_percentile_list[FIO_IO_U_LIST_MAX_LEN];
332
333extern struct thread_data *threads;
334
335static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
336{
337 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
338}
339
340#define BLOCKS_PER_MAP (8 * sizeof(unsigned long))
341#define TO_MAP_BLOCK(f, b) (b)
342#define RAND_MAP_IDX(f, b) (TO_MAP_BLOCK(f, b) / BLOCKS_PER_MAP)
343#define RAND_MAP_BIT(f, b) (TO_MAP_BLOCK(f, b) & (BLOCKS_PER_MAP - 1))
344
345#define REAL_MAX_JOBS 2048
346
347static inline enum error_type td_error_type(enum fio_ddir ddir, int err)
348{
349 if (err == EILSEQ)
350 return ERROR_TYPE_VERIFY_BIT;
351 if (ddir == DDIR_READ)
352 return ERROR_TYPE_READ_BIT;
353 return ERROR_TYPE_WRITE_BIT;
354}
355
356static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
357static inline int td_non_fatal_error(struct thread_data *td,
358 enum error_type_bit etype, int err)
359{
360 int i;
361 if (!td->o.ignore_error[etype]) {
362 td->o.ignore_error[etype] = __NON_FATAL_ERR;
363 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
364 / sizeof(int);
365 }
366
367 if (!(td->o.continue_on_error & (1 << etype)))
368 return 0;
369 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
370 if (td->o.ignore_error[etype][i] == err)
371 return 1;
372 return 0;
373}
374
375static inline void update_error_count(struct thread_data *td, int err)
376{
377 td->total_err_count++;
378 if (td->total_err_count == 1)
379 td->first_error = err;
380}
381
382static inline int should_fsync(struct thread_data *td)
383{
384 if (td->last_was_sync)
385 return 0;
386 if (td->o.odirect)
387 return 0;
388 if (td_write(td) || td_rw(td) || td->o.override_sync)
389 return 1;
390
391 return 0;
392}
393
394/*
395 * Init/option functions
396 */
397extern int __must_check fio_init_options(void);
398extern int __must_check parse_options(int, char **);
399extern int parse_jobs_ini(char *, int, int, int);
400extern int parse_cmd_line(int, char **, int);
401extern int fio_backend(void);
402extern void reset_fio_state(void);
403extern void clear_io_state(struct thread_data *);
404extern int fio_options_parse(struct thread_data *, char **, int);
405extern void fio_keywords_init(void);
406extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
407extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
408extern void fio_fill_default_options(struct thread_data *);
409extern int fio_show_option_help(const char *);
410extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
411extern void fio_options_dup_and_init(struct option *);
412extern void fio_options_mem_dupe(struct thread_data *);
413extern void options_mem_dupe(void *data, struct fio_option *options);
414extern void td_fill_rand_seeds(struct thread_data *);
415extern void add_job_opts(const char **, int);
416extern char *num2str(unsigned long, int, int, int);
417extern int ioengine_load(struct thread_data *);
418
419extern unsigned long page_mask;
420extern unsigned long page_size;
421extern int initialize_fio(char *envp[]);
422
423#define FIO_GETOPT_JOB 0x89000000
424#define FIO_GETOPT_IOENGINE 0x98000000
425#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
426
427/*
428 * ETA/status stuff
429 */
430extern void print_thread_status(void);
431extern void print_status_init(int);
432extern char *fio_uint_to_kmg(unsigned int val);
433
434/*
435 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
436 * will never back again. It may cycle between running/verififying/fsyncing.
437 * Once the thread reaches TD_EXITED, it is just waiting for the core to
438 * reap it.
439 */
440enum {
441 TD_NOT_CREATED = 0,
442 TD_CREATED,
443 TD_INITIALIZED,
444 TD_RAMP,
445 TD_RUNNING,
446 TD_PRE_READING,
447 TD_VERIFYING,
448 TD_FSYNCING,
449 TD_EXITED,
450 TD_REAPED,
451};
452
453extern void td_set_runstate(struct thread_data *, int);
454#define TERMINATE_ALL (-1)
455extern void fio_terminate_threads(int);
456
457/*
458 * Memory helpers
459 */
460extern int __must_check fio_pin_memory(struct thread_data *);
461extern void fio_unpin_memory(struct thread_data *);
462extern int __must_check allocate_io_mem(struct thread_data *);
463extern void free_io_mem(struct thread_data *);
464extern void free_threads_shm(void);
465
466/*
467 * Reset stats after ramp time completes
468 */
469extern void reset_all_stats(struct thread_data *);
470
471/*
472 * blktrace support
473 */
474#ifdef FIO_HAVE_BLKTRACE
475extern int is_blktrace(const char *);
476extern int load_blktrace(struct thread_data *, const char *);
477#endif
478
479/*
480 * Mark unused variables passed to ops functions as unused, to silence gcc
481 */
482#define fio_unused __attribute((__unused__))
483#define fio_init __attribute__((constructor))
484#define fio_exit __attribute__((destructor))
485
486#define for_each_td(td, i) \
487 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
488#define for_each_file(td, f, i) \
489 if ((td)->files_index) \
490 for ((i) = 0, (f) = (td)->files[0]; \
491 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
492 (i)++)
493
494#define fio_assert(td, cond) do { \
495 if (!(cond)) { \
496 int *__foo = NULL; \
497 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
498 td_set_runstate((td), TD_EXITED); \
499 (td)->error = EFAULT; \
500 *__foo = 0; \
501 } \
502} while (0)
503
504static inline int fio_fill_issue_time(struct thread_data *td)
505{
506 if (td->o.read_iolog_file ||
507 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
508 return 1;
509
510 return 0;
511}
512
513static inline int __should_check_rate(struct thread_data *td,
514 enum fio_ddir ddir)
515{
516 struct thread_options *o = &td->o;
517
518 /*
519 * If some rate setting was given, we need to check it
520 */
521 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
522 o->rate_iops_min[ddir])
523 return 1;
524
525 return 0;
526}
527
528static inline int should_check_rate(struct thread_data *td,
529 unsigned long *bytes_done)
530{
531 int ret = 0;
532
533 if (bytes_done[DDIR_READ])
534 ret |= __should_check_rate(td, DDIR_READ);
535 if (bytes_done[DDIR_WRITE])
536 ret |= __should_check_rate(td, DDIR_WRITE);
537 if (bytes_done[DDIR_TRIM])
538 ret |= __should_check_rate(td, DDIR_TRIM);
539
540 return ret;
541}
542
543static inline int is_power_of_2(unsigned int val)
544{
545 return (val != 0 && ((val & (val - 1)) == 0));
546}
547
548/*
549 * We currently only need to do locking if we have verifier threads
550 * accessing our internal structures too
551 */
552static inline void td_io_u_lock(struct thread_data *td)
553{
554 if (td->o.verify_async)
555 pthread_mutex_lock(&td->io_u_lock);
556}
557
558static inline void td_io_u_unlock(struct thread_data *td)
559{
560 if (td->o.verify_async)
561 pthread_mutex_unlock(&td->io_u_lock);
562}
563
564static inline void td_io_u_free_notify(struct thread_data *td)
565{
566 if (td->o.verify_async)
567 pthread_cond_signal(&td->free_cond);
568}
569
570extern const char *fio_get_arch_string(int);
571extern const char *fio_get_os_string(int);
572
573#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
574
575enum {
576 FIO_OUTPUT_TERSE = 0,
577 FIO_OUTPUT_JSON,
578 FIO_OUTPUT_NORMAL,
579};
580
581#endif