Make lockmem a per job option
[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 "arch/arch.h"
24#include "os/os.h"
25#include "mutex.h"
26#include "log.h"
27#include "debug.h"
28#include "file.h"
29#include "io_ddir.h"
30#include "ioengine.h"
31#include "iolog.h"
32#include "helpers.h"
33#include "options.h"
34#include "profile.h"
35#include "time.h"
36#include "gettime.h"
37#include "lib/getopt.h"
38#include "lib/rand.h"
39#include "lib/rbtree.h"
40#include "server.h"
41#include "stat.h"
42#include "flow.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_NR_OFFS,
86};
87
88/*
89 * This describes a single thread/process executing a fio job.
90 */
91struct thread_data {
92 struct thread_options o;
93 unsigned long flags;
94 void *eo;
95 char verror[FIO_VERROR_SIZE];
96 pthread_t thread;
97 int thread_number;
98 int groupid;
99 struct thread_stat ts;
100
101 struct io_log *slat_log;
102 struct io_log *clat_log;
103 struct io_log *lat_log;
104 struct io_log *bw_log;
105 struct io_log *iops_log;
106
107 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
108 struct timeval bw_sample_time;
109
110 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
111 struct timeval iops_sample_time;
112
113 volatile int update_rusage;
114 struct fio_mutex *rusage_sem;
115 struct rusage ru_start;
116 struct rusage ru_end;
117
118 struct fio_file **files;
119 unsigned char *file_locks;
120 unsigned int files_size;
121 unsigned int files_index;
122 unsigned int nr_open_files;
123 unsigned int nr_done_files;
124 unsigned int nr_normal_files;
125 union {
126 unsigned int next_file;
127 os_random_state_t next_file_state;
128 struct frand_state __next_file_state;
129 };
130 int error;
131 int sig;
132 int done;
133 pid_t pid;
134 char *orig_buffer;
135 size_t orig_buffer_size;
136 volatile int terminate;
137 volatile int runstate;
138 unsigned int ioprio;
139 unsigned int ioprio_set;
140 unsigned int last_was_sync;
141 enum fio_ddir last_ddir;
142
143 char *mmapfile;
144 int mmapfd;
145
146 void *iolog_buf;
147 FILE *iolog_f;
148
149 char *sysfs_root;
150
151 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
152
153 union {
154 os_random_state_t bsrange_state;
155 struct frand_state __bsrange_state;
156 };
157 union {
158 os_random_state_t verify_state;
159 struct frand_state __verify_state;
160 };
161 union {
162 os_random_state_t trim_state;
163 struct frand_state __trim_state;
164 };
165
166 struct frand_state buf_state;
167
168 unsigned int verify_batch;
169 unsigned int trim_batch;
170
171 int shm_id;
172
173 /*
174 * IO engine hooks, contains everything needed to submit an io_u
175 * to any of the available IO engines.
176 */
177 struct ioengine_ops *io_ops;
178
179 /*
180 * Queue depth of io_u's that fio MIGHT do
181 */
182 unsigned int cur_depth;
183
184 /*
185 * io_u's about to be committed
186 */
187 unsigned int io_u_queued;
188
189 /*
190 * io_u's submitted but not completed yet
191 */
192 unsigned int io_u_in_flight;
193
194 /*
195 * List of free and busy io_u's
196 */
197 struct flist_head io_u_freelist;
198 struct flist_head io_u_busylist;
199 struct flist_head io_u_requeues;
200 pthread_mutex_t io_u_lock;
201 pthread_cond_t free_cond;
202
203 /*
204 * async verify offload
205 */
206 struct flist_head verify_list;
207 pthread_t *verify_threads;
208 unsigned int nr_verify_threads;
209 pthread_cond_t verify_cond;
210 int verify_thread_exit;
211
212 /*
213 * Rate state
214 */
215 uint64_t rate_bps[DDIR_RWDIR_CNT];
216 long rate_pending_usleep[DDIR_RWDIR_CNT];
217 unsigned long rate_bytes[DDIR_RWDIR_CNT];
218 unsigned long rate_blocks[DDIR_RWDIR_CNT];
219 struct timeval lastrate[DDIR_RWDIR_CNT];
220
221 uint64_t total_io_size;
222 uint64_t fill_device_size;
223
224 unsigned long io_issues[DDIR_RWDIR_CNT];
225 uint64_t io_blocks[DDIR_RWDIR_CNT];
226 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
227 uint64_t io_bytes[DDIR_RWDIR_CNT];
228 uint64_t io_skip_bytes;
229 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
230 uint64_t zone_bytes;
231 struct fio_mutex *mutex;
232
233 /*
234 * State for random io, a bitmap of blocks done vs not done
235 */
236 union {
237 os_random_state_t random_state;
238 struct frand_state __random_state;
239 };
240
241 struct timeval start; /* start of this loop */
242 struct timeval epoch; /* time job was started */
243 struct timeval last_issue;
244 struct timeval tv_cache;
245 unsigned int tv_cache_nr;
246 unsigned int tv_cache_mask;
247 unsigned int ramp_time_over;
248
249 /*
250 * read/write mixed workload state
251 */
252 union {
253 os_random_state_t rwmix_state;
254 struct frand_state __rwmix_state;
255 };
256 unsigned long rwmix_issues;
257 enum fio_ddir rwmix_ddir;
258 unsigned int ddir_seq_nr;
259
260 /*
261 * IO history logs for verification. We use a tree for sorting,
262 * if we are overwriting. Otherwise just use a fifo.
263 */
264 struct rb_root io_hist_tree;
265 struct flist_head io_hist_list;
266 unsigned long io_hist_len;
267
268 /*
269 * For IO replaying
270 */
271 struct flist_head io_log_list;
272
273 /*
274 * For tracking/handling discards
275 */
276 struct flist_head trim_list;
277 unsigned long trim_entries;
278
279 struct flist_head next_rand_list;
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), "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 unsigned int nr_process, nr_thread;
349extern int shm_id;
350extern int groupid;
351extern int output_format;
352extern int temp_stall_ts;
353extern uintptr_t page_mask, page_size;
354extern int read_only;
355extern int eta_print;
356extern int eta_new_line;
357extern unsigned long done_secs;
358extern char *job_section;
359extern int fio_gtod_offload;
360extern int fio_gtod_cpu;
361extern enum fio_cs fio_clock_source;
362extern int fio_clock_source_set;
363extern int warnings_fatal;
364extern int terse_version;
365extern int is_backend;
366extern int nr_clients;
367extern int log_syslog;
368extern const char fio_version_string[];
369
370extern struct thread_data *threads;
371
372static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
373{
374 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
375}
376
377#define REAL_MAX_JOBS 2048
378
379static inline enum error_type_bit td_error_type(enum fio_ddir ddir, int err)
380{
381 if (err == EILSEQ)
382 return ERROR_TYPE_VERIFY_BIT;
383 if (ddir == DDIR_READ)
384 return ERROR_TYPE_READ_BIT;
385 return ERROR_TYPE_WRITE_BIT;
386}
387
388static int __NON_FATAL_ERR[] = {EIO, EILSEQ};
389static inline int td_non_fatal_error(struct thread_data *td,
390 enum error_type_bit etype, int err)
391{
392 int i;
393 if (!td->o.ignore_error[etype]) {
394 td->o.ignore_error[etype] = __NON_FATAL_ERR;
395 td->o.ignore_error_nr[etype] = sizeof(__NON_FATAL_ERR)
396 / sizeof(int);
397 }
398
399 if (!(td->o.continue_on_error & (1 << etype)))
400 return 0;
401 for (i = 0; i < td->o.ignore_error_nr[etype]; i++)
402 if (td->o.ignore_error[etype][i] == err)
403 return 1;
404 return 0;
405}
406
407static inline void update_error_count(struct thread_data *td, int err)
408{
409 td->total_err_count++;
410 if (td->total_err_count == 1)
411 td->first_error = err;
412}
413
414static inline int should_fsync(struct thread_data *td)
415{
416 if (td->last_was_sync)
417 return 0;
418 if (td_write(td) || td_rw(td) || td->o.override_sync)
419 return 1;
420
421 return 0;
422}
423
424/*
425 * Init/option functions
426 */
427extern int __must_check parse_options(int, char **);
428extern int parse_jobs_ini(char *, int, int);
429extern int parse_cmd_line(int, char **);
430extern int fio_backend(void);
431extern void reset_fio_state(void);
432extern void clear_io_state(struct thread_data *);
433extern int fio_options_parse(struct thread_data *, char **, int);
434extern void fio_keywords_init(void);
435extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
436extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
437extern void fio_fill_default_options(struct thread_data *);
438extern int fio_show_option_help(const char *);
439extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
440extern void fio_options_dup_and_init(struct option *);
441extern void fio_options_mem_dupe(struct thread_data *);
442extern void options_mem_dupe(void *data, struct fio_option *options);
443extern void td_fill_rand_seeds(struct thread_data *);
444extern void add_job_opts(const char **);
445extern char *num2str(unsigned long, int, int, int, int);
446extern int ioengine_load(struct thread_data *);
447
448#define FIO_GETOPT_JOB 0x89000000
449#define FIO_GETOPT_IOENGINE 0x98000000
450#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
451
452/*
453 * ETA/status stuff
454 */
455extern void print_thread_status(void);
456extern void print_status_init(int);
457
458/*
459 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
460 * will never back again. It may cycle between running/verififying/fsyncing.
461 * Once the thread reaches TD_EXITED, it is just waiting for the core to
462 * reap it.
463 */
464enum {
465 TD_NOT_CREATED = 0,
466 TD_CREATED,
467 TD_INITIALIZED,
468 TD_RAMP,
469 TD_SETTING_UP,
470 TD_RUNNING,
471 TD_PRE_READING,
472 TD_VERIFYING,
473 TD_FSYNCING,
474 TD_EXITED,
475 TD_REAPED,
476};
477
478extern void td_set_runstate(struct thread_data *, int);
479#define TERMINATE_ALL (-1)
480extern void fio_terminate_threads(int);
481
482/*
483 * Memory helpers
484 */
485extern int __must_check fio_pin_memory(struct thread_data *);
486extern void fio_unpin_memory(struct thread_data *);
487extern int __must_check allocate_io_mem(struct thread_data *);
488extern void free_io_mem(struct thread_data *);
489
490/*
491 * Reset stats after ramp time completes
492 */
493extern void reset_all_stats(struct thread_data *);
494
495/*
496 * blktrace support
497 */
498#ifdef FIO_HAVE_BLKTRACE
499extern int is_blktrace(const char *);
500extern int load_blktrace(struct thread_data *, const char *);
501#endif
502
503/*
504 * Mark unused variables passed to ops functions as unused, to silence gcc
505 */
506#define fio_unused __attribute((__unused__))
507#define fio_init __attribute__((constructor))
508#define fio_exit __attribute__((destructor))
509
510#define for_each_td(td, i) \
511 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
512#define for_each_file(td, f, i) \
513 if ((td)->files_index) \
514 for ((i) = 0, (f) = (td)->files[0]; \
515 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
516 (i)++)
517
518#define fio_assert(td, cond) do { \
519 if (!(cond)) { \
520 int *__foo = NULL; \
521 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
522 td_set_runstate((td), TD_EXITED); \
523 (td)->error = EFAULT; \
524 *__foo = 0; \
525 } \
526} while (0)
527
528static inline int fio_fill_issue_time(struct thread_data *td)
529{
530 if (td->o.read_iolog_file ||
531 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
532 return 1;
533
534 return 0;
535}
536
537static inline int __should_check_rate(struct thread_data *td,
538 enum fio_ddir ddir)
539{
540 struct thread_options *o = &td->o;
541
542 /*
543 * If some rate setting was given, we need to check it
544 */
545 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
546 o->rate_iops_min[ddir])
547 return 1;
548
549 return 0;
550}
551
552static inline int should_check_rate(struct thread_data *td,
553 uint64_t *bytes_done)
554{
555 int ret = 0;
556
557 if (bytes_done[DDIR_READ])
558 ret |= __should_check_rate(td, DDIR_READ);
559 if (bytes_done[DDIR_WRITE])
560 ret |= __should_check_rate(td, DDIR_WRITE);
561 if (bytes_done[DDIR_TRIM])
562 ret |= __should_check_rate(td, DDIR_TRIM);
563
564 return ret;
565}
566
567static inline unsigned int td_max_bs(struct thread_data *td)
568{
569 unsigned int max_bs;
570
571 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
572 return max(td->o.max_bs[DDIR_TRIM], max_bs);
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
605enum {
606 FIO_OUTPUT_TERSE = 0,
607 FIO_OUTPUT_JSON,
608 FIO_OUTPUT_NORMAL,
609};
610
611enum {
612 FIO_RAND_DIST_RANDOM = 0,
613 FIO_RAND_DIST_ZIPF,
614 FIO_RAND_DIST_PARETO,
615};
616
617enum {
618 FIO_RAND_GEN_TAUSWORTHE = 0,
619 FIO_RAND_GEN_LFSR,
620};
621
622#endif