os-linux.h revision fffca02dfd1b700903ba23b67fab7b59bf460ec2 
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
31#define OS_MAP_ANON		(MAP_ANONYMOUS)
32
33#ifndef CLOCK_MONOTONIC
34#define CLOCK_MONOTONIC 1
35#endif
36
37#ifdef FIO_HAVE_CPU_AFFINITY
38typedef cpu_set_t os_cpu_mask_t;
39#else
40typedef int os_cpu_mask_t;
41#endif
42typedef struct drand48_data os_random_state_t;
43
44/*
45 * we want fadvise64 really, but it's so tangled... later
46 */
47#ifdef FIO_HAVE_FADVISE
48#define fadvise(fd, off, len, advice)	\
49	posix_fadvise((fd), (off_t)(off), (len), (advice))
50#endif
51
52/*
53 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
54 * the affinity helpers to work.
55 */
56#ifdef FIO_HAVE_CPU_AFFINITY
57#ifndef GLIBC_2_3_2
58#define fio_setaffinity(td)		\
59	sched_setaffinity((td)->pid, sizeof((td)->o.cpumask), &(td)->o.cpumask)
60#define fio_getaffinity(pid, ptr)	\
61	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
62#else
63#define fio_setaffinity(td)		\
64	sched_setaffinity((td)->pid, &(td)->o.cpumask)
65#define fio_getaffinity(pid, ptr)	\
66	sched_getaffinity((pid), (ptr))
67#endif
68#endif
69
70static inline int ioprio_set(int which, int who, int ioprio)
71{
72	return syscall(__NR_ioprio_set, which, who, ioprio);
73}
74
75/*
76 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
77 * aren't either.
78 */
79#ifndef SPLICE_F_MOVE
80#define SPLICE_F_MOVE	(0x01)	/* move pages instead of copying */
81#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
82				 /* we may still block on the fd we splice */
83				 /* from/to, of course */
84#define SPLICE_F_MORE	(0x04)	/* expect more data */
85#define SPLICE_F_GIFT   (0x08)  /* pages passed in are a gift */
86
87static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
88			 size_t len, unsigned int flags)
89{
90	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
91}
92
93static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
94{
95	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
96}
97
98static inline int vmsplice(int fd, const struct iovec *iov,
99			   unsigned long nr_segs, unsigned int flags)
100{
101	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
102}
103#endif
104
105#define SPLICE_DEF_SIZE	(64*1024)
106
107#ifdef FIO_HAVE_SYSLET
108
109struct syslet_uatom;
110struct async_head_user;
111
112/*
113 * syslet stuff
114 */
115static inline struct syslet_uatom *
116async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
117{
118	return (void *) syscall(__NR_async_exec, atom, ahu);
119}
120
121static inline long
122async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
123	   struct async_head_user *ahu)
124{
125	return syscall(__NR_async_wait, min_wait_events,
126			user_ring_idx, ahu);
127}
128
129static inline long async_thread(void *event, struct async_head_user *ahu)
130{
131	return syscall(__NR_async_thread, event, ahu);
132}
133
134static inline long umem_add(unsigned long *uptr, unsigned long inc)
135{
136	return syscall(__NR_umem_add, uptr, inc);
137}
138#endif /* FIO_HAVE_SYSLET */
139
140enum {
141	IOPRIO_CLASS_NONE,
142	IOPRIO_CLASS_RT,
143	IOPRIO_CLASS_BE,
144	IOPRIO_CLASS_IDLE,
145};
146
147enum {
148	IOPRIO_WHO_PROCESS = 1,
149	IOPRIO_WHO_PGRP,
150	IOPRIO_WHO_USER,
151};
152
153#define IOPRIO_BITS		16
154#define IOPRIO_CLASS_SHIFT	13
155
156#ifndef BLKGETSIZE64
157#define BLKGETSIZE64	_IOR(0x12,114,size_t)
158#endif
159
160#ifndef BLKFLSBUF
161#define BLKFLSBUF	_IO(0x12,97)
162#endif
163
164static inline int blockdev_invalidate_cache(int fd)
165{
166	return ioctl(fd, BLKFLSBUF);
167}
168
169static inline int blockdev_size(int fd, unsigned long long *bytes)
170{
171	if (!ioctl(fd, BLKGETSIZE64, bytes))
172		return 0;
173
174	return errno;
175}
176
177static inline unsigned long long os_phys_mem(void)
178{
179	long pagesize, pages;
180
181	pagesize = sysconf(_SC_PAGESIZE);
182	pages = sysconf(_SC_PHYS_PAGES);
183	if (pages == -1 || pagesize == -1)
184		return 0;
185
186	return (unsigned long long) pages * (unsigned long long) pagesize;
187}
188
189static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
190{
191	srand48_r(seed, rs);
192}
193
194static inline long os_random_long(os_random_state_t *rs)
195{
196	long val;
197
198	lrand48_r(rs, &val);
199	return val;
200}
201
202static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
203{
204	struct raw_config_request rq;
205	int fd;
206
207	if (major(dev) != RAW_MAJOR)
208		return 1;
209
210	/*
211	 * we should be able to find /dev/rawctl or /dev/raw/rawctl
212	 */
213	fd = open("/dev/rawctl", O_RDONLY);
214	if (fd < 0) {
215		fd = open("/dev/raw/rawctl", O_RDONLY);
216		if (fd < 0)
217			return 1;
218	}
219
220	rq.raw_minor = minor(dev);
221	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
222		close(fd);
223		return 1;
224	}
225
226	close(fd);
227	*majdev = rq.block_major;
228	*mindev = rq.block_minor;
229	return 0;
230}
231
232#define FIO_O_NOATIME	O_NOATIME
233
234#endif
235