1#ifndef FIO_OS_LINUX_H
2#define FIO_OS_LINUX_H
3
4#define	FIO_OS	os_linux
5
6#include <sys/ioctl.h>
7#include <sys/uio.h>
8#include <sys/syscall.h>
9#include <sys/vfs.h>
10#include <sys/mman.h>
11#include <unistd.h>
12#include <fcntl.h>
13#include <errno.h>
14#include <sched.h>
15#include <linux/unistd.h>
16#include <linux/raw.h>
17#include <linux/major.h>
18#include <byteswap.h>
19
20#include "binject.h"
21#include "../file.h"
22
23#define FIO_HAVE_CPU_AFFINITY
24#define FIO_HAVE_DISK_UTIL
25#define FIO_HAVE_SGIO
26#define FIO_HAVE_IOPRIO
27#define FIO_HAVE_IOSCHED_SWITCH
28#define FIO_HAVE_ODIRECT
29#define FIO_HAVE_HUGETLB
30#define FIO_HAVE_RAWBIND
31#define FIO_HAVE_BLKTRACE
32#define FIO_HAVE_PSHARED_MUTEX
33#define FIO_HAVE_CL_SIZE
34#define FIO_HAVE_CGROUPS
35#define FIO_HAVE_FS_STAT
36#define FIO_HAVE_TRIM
37#define FIO_HAVE_BINJECT
38#define FIO_HAVE_GETTID
39#define FIO_USE_GENERIC_INIT_RANDOM_STATE
40
41#ifdef MAP_HUGETLB
42#define FIO_HAVE_MMAP_HUGE
43#endif
44
45#define OS_MAP_ANON		MAP_ANONYMOUS
46
47typedef cpu_set_t os_cpu_mask_t;
48
49typedef struct drand48_data os_random_state_t;
50
51#ifdef CONFIG_3ARG_AFFINITY
52#define fio_setaffinity(pid, cpumask)		\
53	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
54#define fio_getaffinity(pid, ptr)	\
55	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
56#elif defined(CONFIG_2ARG_AFFINITY)
57#define fio_setaffinity(pid, cpumask)	\
58	sched_setaffinity((pid), &(cpumask))
59#define fio_getaffinity(pid, ptr)	\
60	sched_getaffinity((pid), (ptr))
61#endif
62
63#define fio_cpu_clear(mask, cpu)	(void) CPU_CLR((cpu), (mask))
64#define fio_cpu_set(mask, cpu)		(void) CPU_SET((cpu), (mask))
65#define fio_cpu_isset(mask, cpu)	CPU_ISSET((cpu), (mask))
66#define fio_cpu_count(mask)		CPU_COUNT((mask))
67
68static inline int fio_cpuset_init(os_cpu_mask_t *mask)
69{
70	CPU_ZERO(mask);
71	return 0;
72}
73
74static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
75{
76	return 0;
77}
78
79#define FIO_MAX_CPUS			CPU_SETSIZE
80
81enum {
82	IOPRIO_CLASS_NONE,
83	IOPRIO_CLASS_RT,
84	IOPRIO_CLASS_BE,
85	IOPRIO_CLASS_IDLE,
86};
87
88enum {
89	IOPRIO_WHO_PROCESS = 1,
90	IOPRIO_WHO_PGRP,
91	IOPRIO_WHO_USER,
92};
93
94#define IOPRIO_BITS		16
95#define IOPRIO_CLASS_SHIFT	13
96
97static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
98{
99	/*
100	 * If no class is set, assume BE
101	 */
102	if (!ioprio_class)
103		ioprio_class = IOPRIO_CLASS_BE;
104
105	ioprio |= ioprio_class << IOPRIO_CLASS_SHIFT;
106	return syscall(__NR_ioprio_set, which, who, ioprio);
107}
108
109static inline int gettid(void)
110{
111	return syscall(__NR_gettid);
112}
113
114#define SPLICE_DEF_SIZE	(64*1024)
115
116#ifndef BLKGETSIZE64
117#define BLKGETSIZE64	_IOR(0x12,114,size_t)
118#endif
119
120#ifndef BLKFLSBUF
121#define BLKFLSBUF	_IO(0x12,97)
122#endif
123
124#ifndef BLKDISCARD
125#define BLKDISCARD	_IO(0x12,119)
126#endif
127
128static inline int blockdev_invalidate_cache(struct fio_file *f)
129{
130	return ioctl(f->fd, BLKFLSBUF);
131}
132
133static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
134{
135	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
136		return 0;
137
138	return errno;
139}
140
141static inline unsigned long long os_phys_mem(void)
142{
143	long pagesize, pages;
144
145	pagesize = sysconf(_SC_PAGESIZE);
146	pages = sysconf(_SC_PHYS_PAGES);
147	if (pages == -1 || pagesize == -1)
148		return 0;
149
150	return (unsigned long long) pages * (unsigned long long) pagesize;
151}
152
153static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
154{
155	srand48_r(seed, rs);
156}
157
158static inline long os_random_long(os_random_state_t *rs)
159{
160	long val;
161
162	lrand48_r(rs, &val);
163	return val;
164}
165
166static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
167{
168	struct raw_config_request rq;
169	int fd;
170
171	if (major(dev) != RAW_MAJOR)
172		return 1;
173
174	/*
175	 * we should be able to find /dev/rawctl or /dev/raw/rawctl
176	 */
177	fd = open("/dev/rawctl", O_RDONLY);
178	if (fd < 0) {
179		fd = open("/dev/raw/rawctl", O_RDONLY);
180		if (fd < 0)
181			return 1;
182	}
183
184	rq.raw_minor = minor(dev);
185	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
186		close(fd);
187		return 1;
188	}
189
190	close(fd);
191	*majdev = rq.block_major;
192	*mindev = rq.block_minor;
193	return 0;
194}
195
196#ifdef O_NOATIME
197#define FIO_O_NOATIME	O_NOATIME
198#else
199#define FIO_O_NOATIME	0
200#endif
201
202#ifdef O_ATOMIC
203#define OS_O_ATOMIC	O_ATOMIC
204#else
205#define OS_O_ATOMIC	040000000
206#endif
207
208#ifdef MADV_REMOVE
209#define FIO_MADV_FREE	MADV_REMOVE
210#endif
211
212#if defined(__builtin_bswap16)
213#define fio_swap16(x)	__builtin_bswap16(x)
214#else
215#define fio_swap16(x)	__bswap_16(x)
216#endif
217#if defined(__builtin_bswap32)
218#define fio_swap32(x)	__builtin_bswap32(x)
219#else
220#define fio_swap32(x)	__bswap_32(x)
221#endif
222#if defined(__builtin_bswap64)
223#define fio_swap64(x)	__builtin_bswap64(x)
224#else
225#define fio_swap64(x)	__bswap_64(x)
226#endif
227
228#define CACHE_LINE_FILE	\
229	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
230
231static inline int arch_cache_line_size(void)
232{
233	char size[32];
234	int fd, ret;
235
236	fd = open(CACHE_LINE_FILE, O_RDONLY);
237	if (fd < 0)
238		return -1;
239
240	ret = read(fd, size, sizeof(size));
241
242	close(fd);
243
244	if (ret <= 0)
245		return -1;
246	else
247		return atoi(size);
248}
249
250static inline unsigned long long get_fs_size(const char *path)
251{
252	unsigned long long ret;
253	struct statfs s;
254
255	if (statfs(path, &s) < 0)
256		return -1ULL;
257
258	ret = s.f_bsize;
259	ret *= (unsigned long long) s.f_bfree;
260	return ret;
261}
262
263static inline int os_trim(int fd, unsigned long long start,
264			  unsigned long long len)
265{
266	uint64_t range[2];
267
268	range[0] = start;
269	range[1] = len;
270
271	if (!ioctl(fd, BLKDISCARD, range))
272		return 0;
273
274	return errno;
275}
276
277#ifdef CONFIG_SCHED_IDLE
278static inline int fio_set_sched_idle(void)
279{
280	struct sched_param p = { .sched_priority = 0, };
281	return sched_setscheduler(gettid(), SCHED_IDLE, &p);
282}
283#endif
284
285#endif
286