os-linux.h revision 4e78e405a12b31291aa88fffc1da2cf43b782ef5 
1#ifndef FIO_OS_LINUX_H
2#define FIO_OS_LINUX_H
3
4#include <sys/ioctl.h>
5#include <sys/uio.h>
6#include <sys/syscall.h>
7#include <unistd.h>
8#include <fcntl.h>
9#include <linux/unistd.h>
10#include <linux/raw.h>
11#include <linux/major.h>
12
13#include "indirect.h"
14
15#define FIO_HAVE_LIBAIO
16#define FIO_HAVE_POSIXAIO
17#define FIO_HAVE_FADVISE
18#define FIO_HAVE_CPU_AFFINITY
19#define FIO_HAVE_DISK_UTIL
20#define FIO_HAVE_SGIO
21#define FIO_HAVE_IOPRIO
22#define FIO_HAVE_SPLICE
23#define FIO_HAVE_IOSCHED_SWITCH
24#define FIO_HAVE_ODIRECT
25#define FIO_HAVE_HUGETLB
26#define FIO_HAVE_RAWBIND
27#define FIO_HAVE_BLKTRACE
28#define FIO_HAVE_STRSEP
29#define FIO_HAVE_FALLOCATE
30#define FIO_HAVE_POSIXAIO_FSYNC
31
32#define OS_MAP_ANON		MAP_ANONYMOUS
33
34#ifndef CLOCK_MONOTONIC
35#define CLOCK_MONOTONIC 1
36#endif
37
38typedef cpu_set_t os_cpu_mask_t;
39
40typedef struct drand48_data os_random_state_t;
41
42/*
43 * we want fadvise64 really, but it's so tangled... later
44 */
45#ifdef FIO_HAVE_FADVISE
46#define fadvise(fd, off, len, advice)	\
47	posix_fadvise((fd), (off_t)(off), (len), (advice))
48#endif
49
50/*
51 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
52 * the affinity helpers to work.
53 */
54#ifndef GLIBC_2_3_2
55#define fio_setaffinity(td)		\
56	sched_setaffinity((td)->pid, sizeof((td)->o.cpumask), &(td)->o.cpumask)
57#define fio_getaffinity(pid, ptr)	\
58	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
59#else
60#define fio_setaffinity(td)		\
61	sched_setaffinity((td)->pid, &(td)->o.cpumask)
62#define fio_getaffinity(pid, ptr)	\
63	sched_getaffinity((pid), (ptr))
64#endif
65
66#define fio_cpu_clear(mask, cpu)	CPU_CLR((cpu), (mask))
67#define fio_cpu_set(mask, cpu)		CPU_SET((cpu), (mask))
68#define fio_cpuset_init(td)		CPU_ZERO(&(td)->o.cpumask)
69#define fio_cpuset_exit(td)		do { } while (0)
70
71#define FIO_MAX_CPUS			CPU_SETSIZE
72
73static inline int ioprio_set(int which, int who, int ioprio)
74{
75	return syscall(__NR_ioprio_set, which, who, ioprio);
76}
77
78/*
79 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
80 * aren't either.
81 */
82#ifndef SPLICE_F_MOVE
83#define SPLICE_F_MOVE	(0x01)	/* move pages instead of copying */
84#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
85				 /* we may still block on the fd we splice */
86				 /* from/to, of course */
87#define SPLICE_F_MORE	(0x04)	/* expect more data */
88#define SPLICE_F_GIFT   (0x08)  /* pages passed in are a gift */
89
90static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
91			 size_t len, unsigned int flags)
92{
93	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
94}
95
96static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
97{
98	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
99}
100
101static inline int vmsplice(int fd, const struct iovec *iov,
102			   unsigned long nr_segs, unsigned int flags)
103{
104	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
105}
106#endif
107
108#define SPLICE_DEF_SIZE	(64*1024)
109
110#ifdef FIO_HAVE_SYSLET
111
112struct syslet_uatom;
113struct async_head_user;
114
115/*
116 * syslet stuff
117 */
118static inline struct syslet_uatom *
119async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
120{
121	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
122}
123
124static inline long
125async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
126	   struct async_head_user *ahu)
127{
128	return syscall(__NR_async_wait, min_wait_events,
129			user_ring_idx, ahu);
130}
131
132static inline long async_thread(void *event, struct async_head_user *ahu)
133{
134	return syscall(__NR_async_thread, event, ahu);
135}
136
137static inline long umem_add(unsigned long *uptr, unsigned long inc)
138{
139	return syscall(__NR_umem_add, uptr, inc);
140}
141#endif /* FIO_HAVE_SYSLET */
142
143enum {
144	IOPRIO_CLASS_NONE,
145	IOPRIO_CLASS_RT,
146	IOPRIO_CLASS_BE,
147	IOPRIO_CLASS_IDLE,
148};
149
150enum {
151	IOPRIO_WHO_PROCESS = 1,
152	IOPRIO_WHO_PGRP,
153	IOPRIO_WHO_USER,
154};
155
156#define IOPRIO_BITS		16
157#define IOPRIO_CLASS_SHIFT	13
158
159#ifndef BLKGETSIZE64
160#define BLKGETSIZE64	_IOR(0x12,114,size_t)
161#endif
162
163#ifndef BLKFLSBUF
164#define BLKFLSBUF	_IO(0x12,97)
165#endif
166
167static inline int blockdev_invalidate_cache(int fd)
168{
169	return ioctl(fd, BLKFLSBUF);
170}
171
172static inline int blockdev_size(int fd, unsigned long long *bytes)
173{
174	if (!ioctl(fd, BLKGETSIZE64, bytes))
175		return 0;
176
177	return errno;
178}
179
180static inline unsigned long long os_phys_mem(void)
181{
182	long pagesize, pages;
183
184	pagesize = sysconf(_SC_PAGESIZE);
185	pages = sysconf(_SC_PHYS_PAGES);
186	if (pages == -1 || pagesize == -1)
187		return 0;
188
189	return (unsigned long long) pages * (unsigned long long) pagesize;
190}
191
192static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
193{
194	srand48_r(seed, rs);
195}
196
197static inline long os_random_long(os_random_state_t *rs)
198{
199	long val;
200
201	lrand48_r(rs, &val);
202	return val;
203}
204
205static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
206{
207	struct raw_config_request rq;
208	int fd;
209
210	if (major(dev) != RAW_MAJOR)
211		return 1;
212
213	/*
214	 * we should be able to find /dev/rawctl or /dev/raw/rawctl
215	 */
216	fd = open("/dev/rawctl", O_RDONLY);
217	if (fd < 0) {
218		fd = open("/dev/raw/rawctl", O_RDONLY);
219		if (fd < 0)
220			return 1;
221	}
222
223	rq.raw_minor = minor(dev);
224	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
225		close(fd);
226		return 1;
227	}
228
229	close(fd);
230	*majdev = rq.block_major;
231	*mindev = rq.block_minor;
232	return 0;
233}
234
235#ifdef O_NOATIME
236#define FIO_O_NOATIME	O_NOATIME
237#else
238#define FIO_O_NOATIME	0
239#endif
240
241#endif
242