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 "fio_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	TD_F_COMPRESS		= 128,
75	TD_F_NOIO		= 256,
76};
77
78enum {
79	FIO_RAND_BS_OFF		= 0,
80	FIO_RAND_VER_OFF,
81	FIO_RAND_MIX_OFF,
82	FIO_RAND_FILE_OFF,
83	FIO_RAND_BLOCK_OFF,
84	FIO_RAND_FILE_SIZE_OFF,
85	FIO_RAND_TRIM_OFF,
86	FIO_RAND_BUF_OFF,
87	FIO_RAND_SEQ_RAND_READ_OFF,
88	FIO_RAND_SEQ_RAND_WRITE_OFF,
89	FIO_RAND_SEQ_RAND_TRIM_OFF,
90	FIO_RAND_START_DELAY,
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	volatile int update_rusage;
122	struct fio_mutex *rusage_sem;
123	struct rusage ru_start;
124	struct rusage ru_end;
125
126	struct fio_file **files;
127	unsigned char *file_locks;
128	unsigned int files_size;
129	unsigned int files_index;
130	unsigned int nr_open_files;
131	unsigned int nr_done_files;
132	unsigned int nr_normal_files;
133	union {
134		unsigned int next_file;
135		os_random_state_t next_file_state;
136		struct frand_state __next_file_state;
137	};
138	int error;
139	int sig;
140	int done;
141	pid_t pid;
142	char *orig_buffer;
143	size_t orig_buffer_size;
144	volatile int terminate;
145	volatile int runstate;
146	unsigned int last_was_sync;
147	enum fio_ddir last_ddir;
148
149	int mmapfd;
150
151	void *iolog_buf;
152	FILE *iolog_f;
153
154	char *sysfs_root;
155
156	unsigned long rand_seeds[FIO_RAND_NR_OFFS];
157
158	union {
159		os_random_state_t bsrange_state;
160		struct frand_state __bsrange_state;
161	};
162	union {
163		os_random_state_t verify_state;
164		struct frand_state __verify_state;
165	};
166	union {
167		os_random_state_t trim_state;
168		struct frand_state __trim_state;
169	};
170	union {
171		os_random_state_t delay_state;
172		struct frand_state __delay_state;
173	};
174
175	struct frand_state buf_state;
176
177	unsigned int verify_batch;
178	unsigned int trim_batch;
179
180	int shm_id;
181
182	/*
183	 * IO engine hooks, contains everything needed to submit an io_u
184	 * to any of the available IO engines.
185	 */
186	struct ioengine_ops *io_ops;
187
188	/*
189	 * Queue depth of io_u's that fio MIGHT do
190	 */
191	unsigned int cur_depth;
192
193	/*
194	 * io_u's about to be committed
195	 */
196	unsigned int io_u_queued;
197
198	/*
199	 * io_u's submitted but not completed yet
200	 */
201	unsigned int io_u_in_flight;
202
203	/*
204	 * List of free and busy io_u's
205	 */
206	struct io_u_ring io_u_requeues;
207	struct io_u_queue io_u_freelist;
208	struct io_u_queue io_u_all;
209	pthread_mutex_t io_u_lock;
210	pthread_cond_t free_cond;
211
212	/*
213	 * async verify offload
214	 */
215	struct flist_head verify_list;
216	pthread_t *verify_threads;
217	unsigned int nr_verify_threads;
218	pthread_cond_t verify_cond;
219	int verify_thread_exit;
220
221	/*
222	 * Rate state
223	 */
224	uint64_t rate_bps[DDIR_RWDIR_CNT];
225	long rate_pending_usleep[DDIR_RWDIR_CNT];
226	unsigned long rate_bytes[DDIR_RWDIR_CNT];
227	unsigned long rate_blocks[DDIR_RWDIR_CNT];
228	struct timeval lastrate[DDIR_RWDIR_CNT];
229
230	uint64_t total_io_size;
231	uint64_t fill_device_size;
232
233	unsigned long io_issues[DDIR_RWDIR_CNT];
234	uint64_t io_blocks[DDIR_RWDIR_CNT];
235	uint64_t this_io_blocks[DDIR_RWDIR_CNT];
236	uint64_t io_bytes[DDIR_RWDIR_CNT];
237	uint64_t io_skip_bytes;
238	uint64_t this_io_bytes[DDIR_RWDIR_CNT];
239	uint64_t zone_bytes;
240	struct fio_mutex *mutex;
241
242	/*
243	 * State for random io, a bitmap of blocks done vs not done
244	 */
245	union {
246		os_random_state_t random_state;
247		struct frand_state __random_state;
248	};
249
250	struct timeval start;	/* start of this loop */
251	struct timeval epoch;	/* time job was started */
252	struct timeval last_issue;
253	struct timeval tv_cache;
254	unsigned int tv_cache_nr;
255	unsigned int tv_cache_mask;
256	unsigned int ramp_time_over;
257
258	/*
259	 * Time since last latency_window was started
260	 */
261	struct timeval latency_ts;
262	unsigned int latency_qd;
263	unsigned int latency_qd_high;
264	unsigned int latency_qd_low;
265	unsigned int latency_failed;
266	uint64_t latency_ios;
267	int latency_end_run;
268
269	/*
270	 * read/write mixed workload state
271	 */
272	union {
273		os_random_state_t rwmix_state;
274		struct frand_state __rwmix_state;
275	};
276	unsigned long rwmix_issues;
277	enum fio_ddir rwmix_ddir;
278	unsigned int ddir_seq_nr;
279
280	/*
281	 * rand/seq mixed workload state
282	 */
283	union {
284		os_random_state_t seq_rand_state[DDIR_RWDIR_CNT];
285		struct frand_state __seq_rand_state[DDIR_RWDIR_CNT];
286	};
287
288	/*
289	 * IO history logs for verification. We use a tree for sorting,
290	 * if we are overwriting. Otherwise just use a fifo.
291	 */
292	struct rb_root io_hist_tree;
293	struct flist_head io_hist_list;
294	unsigned long io_hist_len;
295
296	/*
297	 * For IO replaying
298	 */
299	struct flist_head io_log_list;
300
301	/*
302	 * For tracking/handling discards
303	 */
304	struct flist_head trim_list;
305	unsigned long trim_entries;
306
307	struct flist_head next_rand_list;
308
309	/*
310	 * for fileservice, how often to switch to a new file
311	 */
312	unsigned int file_service_nr;
313	unsigned int file_service_left;
314	struct fio_file *file_service_file;
315
316	unsigned int sync_file_range_nr;
317
318	/*
319	 * For generating file sizes
320	 */
321	union {
322		os_random_state_t file_size_state;
323		struct frand_state __file_size_state;
324	};
325
326	/*
327	 * Error counts
328	 */
329	unsigned int total_err_count;
330	int first_error;
331
332	struct fio_flow *flow;
333
334	/*
335	 * Can be overloaded by profiles
336	 */
337	struct prof_io_ops prof_io_ops;
338	void *prof_data;
339
340	void *pinned_mem;
341};
342
343/*
344 * when should interactive ETA output be generated
345 */
346enum {
347	FIO_ETA_AUTO,
348	FIO_ETA_ALWAYS,
349	FIO_ETA_NEVER,
350};
351
352#define __td_verror(td, err, msg, func)					\
353	do {								\
354		unsigned int ____e = (err);				\
355		if ((td)->error)					\
356			break;						\
357		(td)->error = ____e;					\
358		if (!(td)->first_error)					\
359			snprintf(td->verror, sizeof(td->verror), "file:%s:%d, func=%s, error=%s", __FILE__, __LINE__, (func), (msg));		\
360	} while (0)
361
362
363#define td_clear_error(td)		\
364	(td)->error = 0;
365#define td_verror(td, err, func)	\
366	__td_verror((td), (err), strerror((err)), (func))
367#define td_vmsg(td, err, msg, func)	\
368	__td_verror((td), (err), (msg), (func))
369
370#define __fio_stringify_1(x)	#x
371#define __fio_stringify(x)	__fio_stringify_1(x)
372
373extern int exitall_on_terminate;
374extern unsigned int thread_number;
375extern unsigned int stat_number;
376extern int shm_id;
377extern int groupid;
378extern int output_format;
379extern int append_terse_output;
380extern int temp_stall_ts;
381extern uintptr_t page_mask, page_size;
382extern int read_only;
383extern int eta_print;
384extern int eta_new_line;
385extern unsigned long done_secs;
386extern char *job_section;
387extern int fio_gtod_offload;
388extern int fio_gtod_cpu;
389extern enum fio_cs fio_clock_source;
390extern int fio_clock_source_set;
391extern int warnings_fatal;
392extern int terse_version;
393extern int is_backend;
394extern int nr_clients;
395extern int log_syslog;
396extern int status_interval;
397extern const char fio_version_string[];
398
399extern struct thread_data *threads;
400
401static inline void fio_ro_check(struct thread_data *td, struct io_u *io_u)
402{
403	assert(!(io_u->ddir == DDIR_WRITE && !td_write(td)));
404}
405
406#define REAL_MAX_JOBS		2048
407
408static inline int should_fsync(struct thread_data *td)
409{
410	if (td->last_was_sync)
411		return 0;
412	if (td_write(td) || td_rw(td) || td->o.override_sync)
413		return 1;
414
415	return 0;
416}
417
418/*
419 * Init/option functions
420 */
421extern int __must_check fio_init_options(void);
422extern int __must_check parse_options(int, char **);
423extern int parse_jobs_ini(char *, int, int, int);
424extern int parse_cmd_line(int, char **, int);
425extern int fio_backend(void);
426extern void reset_fio_state(void);
427extern void clear_io_state(struct thread_data *);
428extern int fio_options_parse(struct thread_data *, char **, int, int);
429extern void fio_keywords_init(void);
430extern int fio_cmd_option_parse(struct thread_data *, const char *, char *);
431extern int fio_cmd_ioengine_option_parse(struct thread_data *, const char *, char *);
432extern void fio_fill_default_options(struct thread_data *);
433extern int fio_show_option_help(const char *);
434extern void fio_options_set_ioengine_opts(struct option *long_options, struct thread_data *td);
435extern void fio_options_dup_and_init(struct option *);
436extern void fio_options_mem_dupe(struct thread_data *);
437extern void options_mem_dupe(void *data, struct fio_option *options);
438extern void td_fill_rand_seeds(struct thread_data *);
439extern void add_job_opts(const char **, int);
440extern char *num2str(unsigned long, int, int, int, int);
441extern int ioengine_load(struct thread_data *);
442extern int parse_dryrun(void);
443extern int fio_running_or_pending_io_threads(void);
444extern int fio_set_fd_nonblocking(int, const char *);
445
446extern uintptr_t page_mask;
447extern uintptr_t page_size;
448extern int initialize_fio(char *envp[]);
449
450#define FIO_GETOPT_JOB		0x89000000
451#define FIO_GETOPT_IOENGINE	0x98000000
452#define FIO_NR_OPTIONS		(FIO_MAX_OPTS + 128)
453
454/*
455 * ETA/status stuff
456 */
457extern void print_thread_status(void);
458extern void print_status_init(int);
459extern char *fio_uint_to_kmg(unsigned int val);
460
461/*
462 * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
463 * will never back again. It may cycle between running/verififying/fsyncing.
464 * Once the thread reaches TD_EXITED, it is just waiting for the core to
465 * reap it.
466 */
467enum {
468	TD_NOT_CREATED = 0,
469	TD_CREATED,
470	TD_INITIALIZED,
471	TD_RAMP,
472	TD_SETTING_UP,
473	TD_RUNNING,
474	TD_PRE_READING,
475	TD_VERIFYING,
476	TD_FSYNCING,
477	TD_FINISHING,
478	TD_EXITED,
479	TD_REAPED,
480};
481
482extern void td_set_runstate(struct thread_data *, int);
483extern int td_bump_runstate(struct thread_data *, int);
484extern void td_restore_runstate(struct thread_data *, int);
485
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 *, int *);
508extern int load_blktrace(struct thread_data *, const char *, int);
509#endif
510
511/*
512 * Latency target helpers
513 */
514extern void lat_target_check(struct thread_data *);
515extern void lat_target_init(struct thread_data *);
516extern void lat_target_reset(struct thread_data *);
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				    uint64_t *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 unsigned int td_max_bs(struct thread_data *td)
576{
577	unsigned int max_bs;
578
579	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
580	return max(td->o.max_bs[DDIR_TRIM], max_bs);
581}
582
583static inline unsigned int td_min_bs(struct thread_data *td)
584{
585	unsigned int min_bs;
586
587	min_bs = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
588	return min(td->o.min_bs[DDIR_TRIM], min_bs);
589}
590
591static inline int is_power_of_2(unsigned long val)
592{
593	return (val != 0 && ((val & (val - 1)) == 0));
594}
595
596/*
597 * We currently only need to do locking if we have verifier threads
598 * accessing our internal structures too
599 */
600static inline void td_io_u_lock(struct thread_data *td)
601{
602	if (td->o.verify_async)
603		pthread_mutex_lock(&td->io_u_lock);
604}
605
606static inline void td_io_u_unlock(struct thread_data *td)
607{
608	if (td->o.verify_async)
609		pthread_mutex_unlock(&td->io_u_lock);
610}
611
612static inline void td_io_u_free_notify(struct thread_data *td)
613{
614	if (td->o.verify_async)
615		pthread_cond_signal(&td->free_cond);
616}
617
618extern const char *fio_get_arch_string(int);
619extern const char *fio_get_os_string(int);
620
621#ifdef FIO_INTERNAL
622#define ARRAY_SIZE(x) (sizeof((x)) / (sizeof((x)[0])))
623#endif
624
625enum {
626	FIO_OUTPUT_TERSE	= 0,
627	FIO_OUTPUT_JSON,
628	FIO_OUTPUT_NORMAL,
629};
630
631enum {
632	FIO_RAND_DIST_RANDOM	= 0,
633	FIO_RAND_DIST_ZIPF,
634	FIO_RAND_DIST_PARETO,
635};
636
637enum {
638	FIO_RAND_GEN_TAUSWORTHE = 0,
639	FIO_RAND_GEN_LFSR,
640};
641
642enum {
643	FIO_CPUS_SHARED		= 0,
644	FIO_CPUS_SPLIT,
645};
646
647#endif
648