fix memory allocation in the cases we may need to align
[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
43#ifdef CONFIG_SOLARISAIO
44#include <sys/asynch.h>
45#endif
46
47#ifdef CONFIG_LIBNUMA
48#include <linux/mempolicy.h>
49#include <numa.h>
50
51/*
52 * "local" is pseudo-policy
53 */
54#define MPOL_LOCAL MPOL_MAX
55#endif
56
57/*
58 * offset generator types
59 */
60enum {
61 RW_SEQ_SEQ = 0,
62 RW_SEQ_IDENT,
63};
64
65enum {
66 TD_F_VER_BACKLOG = 1,
67 TD_F_TRIM_BACKLOG = 2,
68 TD_F_READ_IOLOG = 4,
69 TD_F_REFILL_BUFFERS = 8,
70 TD_F_SCRAMBLE_BUFFERS = 16,
71 TD_F_VER_NONE = 32,
72 TD_F_PROFILE_OPS = 64,
73};
74
75enum {
76 FIO_RAND_BS_OFF = 0,
77 FIO_RAND_VER_OFF,
78 FIO_RAND_MIX_OFF,
79 FIO_RAND_FILE_OFF,
80 FIO_RAND_BLOCK_OFF,
81 FIO_RAND_FILE_SIZE_OFF,
82 FIO_RAND_TRIM_OFF,
83 FIO_RAND_BUF_OFF,
84 FIO_RAND_NR_OFFS,
85};
86
87/*
88 * This describes a single thread/process executing a fio job.
89 */
90struct thread_data {
91 struct thread_options o;
92 unsigned long flags;
93 void *eo;
94 char verror[FIO_VERROR_SIZE];
95 pthread_t thread;
96 unsigned int thread_number;
97 unsigned int groupid;
98 struct thread_stat ts;
99
100 int client_type;
101
102 struct io_log *slat_log;
103 struct io_log *clat_log;
104 struct io_log *lat_log;
105 struct io_log *bw_log;
106 struct io_log *iops_log;
107
108 uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
109 struct timeval bw_sample_time;
110
111 uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
112 struct timeval iops_sample_time;
113
114 volatile int update_rusage;
115 struct fio_mutex *rusage_sem;
116 struct rusage ru_start;
117 struct rusage ru_end;
118
119 struct fio_file **files;
120 unsigned char *file_locks;
121 unsigned int files_size;
122 unsigned int files_index;
123 unsigned int nr_open_files;
124 unsigned int nr_done_files;
125 unsigned int nr_normal_files;
126 union {
127 unsigned int next_file;
128 os_random_state_t next_file_state;
129 struct frand_state __next_file_state;
130 };
131 int error;
132 int sig;
133 int done;
134 pid_t pid;
135 char *orig_buffer;
136 size_t orig_buffer_size;
137 volatile int terminate;
138 volatile int runstate;
139 unsigned int last_was_sync;
140 enum fio_ddir last_ddir;
141
142 int mmapfd;
143
144 void *iolog_buf;
145 FILE *iolog_f;
146
147 char *sysfs_root;
148
149 unsigned long rand_seeds[FIO_RAND_NR_OFFS];
150
151 union {
152 os_random_state_t bsrange_state;
153 struct frand_state __bsrange_state;
154 };
155 union {
156 os_random_state_t verify_state;
157 struct frand_state __verify_state;
158 };
159 union {
160 os_random_state_t trim_state;
161 struct frand_state __trim_state;
162 };
163
164 struct frand_state buf_state;
165
166 unsigned int verify_batch;
167 unsigned int trim_batch;
168
169 int shm_id;
170
171 /*
172 * IO engine hooks, contains everything needed to submit an io_u
173 * to any of the available IO engines.
174 */
175 struct ioengine_ops *io_ops;
176
177 /*
178 * Queue depth of io_u's that fio MIGHT do
179 */
180 unsigned int cur_depth;
181
182 /*
183 * io_u's about to be committed
184 */
185 unsigned int io_u_queued;
186
187 /*
188 * io_u's submitted but not completed yet
189 */
190 unsigned int io_u_in_flight;
191
192 /*
193 * List of free and busy io_u's
194 */
195 struct flist_head io_u_freelist;
196 struct flist_head io_u_busylist;
197 struct flist_head io_u_requeues;
198 pthread_mutex_t io_u_lock;
199 pthread_cond_t free_cond;
200
201 /*
202 * async verify offload
203 */
204 struct flist_head verify_list;
205 pthread_t *verify_threads;
206 unsigned int nr_verify_threads;
207 pthread_cond_t verify_cond;
208 int verify_thread_exit;
209
210 /*
211 * Rate state
212 */
213 uint64_t rate_bps[DDIR_RWDIR_CNT];
214 long rate_pending_usleep[DDIR_RWDIR_CNT];
215 unsigned long rate_bytes[DDIR_RWDIR_CNT];
216 unsigned long rate_blocks[DDIR_RWDIR_CNT];
217 struct timeval lastrate[DDIR_RWDIR_CNT];
218
219 uint64_t total_io_size;
220 uint64_t fill_device_size;
221
222 unsigned long io_issues[DDIR_RWDIR_CNT];
223 uint64_t io_blocks[DDIR_RWDIR_CNT];
224 uint64_t this_io_blocks[DDIR_RWDIR_CNT];
225 uint64_t io_bytes[DDIR_RWDIR_CNT];
226 uint64_t io_skip_bytes;
227 uint64_t this_io_bytes[DDIR_RWDIR_CNT];
228 uint64_t zone_bytes;
229 struct fio_mutex *mutex;
230
231 /*
232 * State for random io, a bitmap of blocks done vs not done
233 */
234 union {
235 os_random_state_t random_state;
236 struct frand_state __random_state;
237 };
238
239 struct timeval start; /* start of this loop */
240 struct timeval epoch; /* time job was started */
241 struct timeval last_issue;
242 struct timeval tv_cache;
243 unsigned int tv_cache_nr;
244 unsigned int tv_cache_mask;
245 unsigned int ramp_time_over;
246
247 /*
248 * read/write mixed workload state
249 */
250 union {
251 os_random_state_t rwmix_state;
252 struct frand_state __rwmix_state;
253 };
254 unsigned long rwmix_issues;
255 enum fio_ddir rwmix_ddir;
256 unsigned int ddir_seq_nr;
257
258 /*
259 * IO history logs for verification. We use a tree for sorting,
260 * if we are overwriting. Otherwise just use a fifo.
261 */
262 struct rb_root io_hist_tree;
263 struct flist_head io_hist_list;
264 unsigned long io_hist_len;
265
266 /*
267 * For IO replaying
268 */
269 struct flist_head io_log_list;
270
271 /*
272 * For tracking/handling discards
273 */
274 struct flist_head trim_list;
275 unsigned long trim_entries;
276
277 struct flist_head next_rand_list;
278
279 /*
280 * for fileservice, how often to switch to a new file
281 */
282 unsigned int file_service_nr;
283 unsigned int file_service_left;
284 struct fio_file *file_service_file;
285
286 unsigned int sync_file_range_nr;
287
288 /*
289 * For generating file sizes
290 */
291 union {
292 os_random_state_t file_size_state;
293 struct frand_state __file_size_state;
294 };
295
296 /*
297 * Error counts
298 */
299 unsigned int total_err_count;
300 int first_error;
301
302 struct fio_flow *flow;
303
304 /*
305 * Can be overloaded by profiles
306 */
307 struct prof_io_ops prof_io_ops;
308 void *prof_data;
309
310 void *pinned_mem;
311};
312
313/*
314 * when should interactive ETA output be generated
315 */
316enum {
317 FIO_ETA_AUTO,
318 FIO_ETA_ALWAYS,
319 FIO_ETA_NEVER,
320};
321
322#define __td_verror(td, err, msg, func) \
323 do { \
324 int e = (err); \
325 if ((td)->error) \
326 break; \
327 (td)->error = e; \
328 if (!(td)->first_error) \
329 snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg)); \
330 } while (0)
331
332
333#define td_clear_error(td) \
334 (td)->error = 0;
335#define td_verror(td, err, func) \
336 __td_verror((td), (err), strerror((err)), (func))
337#define td_vmsg(td, err, msg, func) \
338 __td_verror((td), (err), (msg), (func))
339
340#define __fio_stringify_1(x) #x
341#define __fio_stringify(x) __fio_stringify_1(x)
342
343extern int exitall_on_terminate;
344extern unsigned int thread_number;
345extern unsigned int stat_number;
346extern int shm_id;
347extern int groupid;
348extern int output_format;
349extern int temp_stall_ts;
350extern uintptr_t page_mask, page_size;
351extern int read_only;
352extern int eta_print;
353extern int eta_new_line;
354extern unsigned long done_secs;
355extern char *job_section;
356extern int fio_gtod_offload;
357extern int fio_gtod_cpu;
358extern enum fio_cs fio_clock_source;
359extern int fio_clock_source_set;
360extern int warnings_fatal;
361extern int terse_version;
362extern int is_backend;
363extern int nr_clients;
364extern int log_syslog;
365extern const char fio_version_string[];
366
367extern struct thread_data *threads;
368
369static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
370{
371 assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
372}
373
374#define REAL_MAX_JOBS 2048
375
376static inline int should_fsync(struct thread_data *td)
377{
378 if (td->last_was_sync)
379 return 0;
380 if (td_write(td) || td_rw(td) || td->o.override_sync)
381 return 1;
382
383 return 0;
384}
385
386/*
387 * Init/option functions
388 */
389extern int __must_check fio_init_options(void);
390extern int __must_check parse_options(int, char **);
391extern int parse_jobs_ini(char *, int, int, int);
392extern int parse_cmd_line(int, char **, int);
393extern int fio_backend(void);
394extern void reset_fio_state(void);
395extern void clear_io_state(struct thread_data *);
396extern int fio_options_parse(struct thread_data *, char **, int);
397extern void fio_keywords_init(void);
398extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
399extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
400extern void fio_fill_default_options(struct thread_data *);
401extern int fio_show_option_help(const char *);
402extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
403extern void fio_options_dup_and_init(struct option *);
404extern void fio_options_mem_dupe(struct thread_data *);
405extern void options_mem_dupe(void *data, struct fio_option *options);
406extern void td_fill_rand_seeds(struct thread_data *);
407extern void add_job_opts(const char **, int);
408extern char *num2str(unsigned long, int, int, int, int);
409extern int ioengine_load(struct thread_data *);
410
411extern uintptr_t page_mask;
412extern uintptr_t page_size;
413extern int initialize_fio(char *envp[]);
414
415#define FIO_GETOPT_JOB 0x89000000
416#define FIO_GETOPT_IOENGINE 0x98000000
417#define FIO_NR_OPTIONS (FIO_MAX_OPTS + 128)
418
419/*
420 * ETA/status stuff
421 */
422extern void print_thread_status(void);
423extern void print_status_init(int);
424extern char *fio_uint_to_kmg(unsigned int val);
425
426/*
427 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
428 * will never back again. It may cycle between running/verififying/fsyncing.
429 * Once the thread reaches TD_EXITED, it is just waiting for the core to
430 * reap it.
431 */
432enum {
433 TD_NOT_CREATED = 0,
434 TD_CREATED,
435 TD_INITIALIZED,
436 TD_RAMP,
437 TD_SETTING_UP,
438 TD_RUNNING,
439 TD_PRE_READING,
440 TD_VERIFYING,
441 TD_FSYNCING,
442 TD_EXITED,
443 TD_REAPED,
444};
445
446extern void td_set_runstate(struct thread_data *, int);
447#define TERMINATE_ALL (-1)
448extern void fio_terminate_threads(int);
449
450/*
451 * Memory helpers
452 */
453extern int __must_check fio_pin_memory(struct thread_data *);
454extern void fio_unpin_memory(struct thread_data *);
455extern int __must_check allocate_io_mem(struct thread_data *);
456extern void free_io_mem(struct thread_data *);
457extern void free_threads_shm(void);
458
459/*
460 * Reset stats after ramp time completes
461 */
462extern void reset_all_stats(struct thread_data *);
463
464/*
465 * blktrace support
466 */
467#ifdef FIO_HAVE_BLKTRACE
468extern int is_blktrace(const char *);
469extern int load_blktrace(struct thread_data *, const char *);
470#endif
471
472#define for_each_td(td, i) \
473 for ((i) = 0, (td) = &threads[0]; (i) < (int) thread_number; (i)++, (td)++)
474#define for_each_file(td, f, i) \
475 if ((td)->files_index) \
476 for ((i) = 0, (f) = (td)->files[0]; \
477 (i) < (td)->o.nr_files && ((f) = (td)->files[i]) != NULL; \
478 (i)++)
479
480#define fio_assert(td, cond) do { \
481 if (!(cond)) { \
482 int *__foo = NULL; \
483 fprintf(stderr, "file:%s:%d, assert %s failed\n", __FILE__, __LINE__, #cond); \
484 td_set_runstate((td), TD_EXITED); \
485 (td)->error = EFAULT; \
486 *__foo = 0; \
487 } \
488} while (0)
489
490static inline int fio_fill_issue_time(struct thread_data *td)
491{
492 if (td->o.read_iolog_file ||
493 !td->o.disable_clat || !td->o.disable_slat || !td->o.disable_bw)
494 return 1;
495
496 return 0;
497}
498
499static inline int __should_check_rate(struct thread_data *td,
500 enum fio_ddir ddir)
501{
502 struct thread_options *o = &td->o;
503
504 /*
505 * If some rate setting was given, we need to check it
506 */
507 if (o->rate[ddir] || o->ratemin[ddir] || o->rate_iops[ddir] ||
508 o->rate_iops_min[ddir])
509 return 1;
510
511 return 0;
512}
513
514static inline int should_check_rate(struct thread_data *td,
515 uint64_t *bytes_done)
516{
517 int ret = 0;
518
519 if (bytes_done[DDIR_READ])
520 ret |= __should_check_rate(td, DDIR_READ);
521 if (bytes_done[DDIR_WRITE])
522 ret |= __should_check_rate(td, DDIR_WRITE);
523 if (bytes_done[DDIR_TRIM])
524 ret |= __should_check_rate(td, DDIR_TRIM);
525
526 return ret;
527}
528
529static inline unsigned int td_max_bs(struct thread_data *td)
530{
531 unsigned int max_bs;
532
533 max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
534 return max(td->o.max_bs[DDIR_TRIM], max_bs);
535}
536
537static inline int is_power_of_2(unsigned int val)
538{
539 return (val != 0 && ((val & (val - 1)) == 0));
540}
541
542/*
543 * We currently only need to do locking if we have verifier threads
544 * accessing our internal structures too
545 */
546static inline void td_io_u_lock(struct thread_data *td)
547{
548 if (td->o.verify_async)
549 pthread_mutex_lock(&td->io_u_lock);
550}
551
552static inline void td_io_u_unlock(struct thread_data *td)
553{
554 if (td->o.verify_async)
555 pthread_mutex_unlock(&td->io_u_lock);
556}
557
558static inline void td_io_u_free_notify(struct thread_data *td)
559{
560 if (td->o.verify_async)
561 pthread_cond_signal(&td->free_cond);
562}
563
564extern const char *fio_get_arch_string(int);
565extern const char *fio_get_os_string(int);
566
567#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
568
569enum {
570 FIO_OUTPUT_TERSE = 0,
571 FIO_OUTPUT_JSON,
572 FIO_OUTPUT_NORMAL,
573};
574
575enum {
576 FIO_RAND_DIST_RANDOM = 0,
577 FIO_RAND_DIST_ZIPF,
578 FIO_RAND_DIST_PARETO,
579};
580
581enum {
582 FIO_RAND_GEN_TAUSWORTHE = 0,
583 FIO_RAND_GEN_LFSR,
584};
585
586#endif