os-linux.h revision a5f3027cb0495dfe217b2626d248fcc054e7e878 
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 <sys/vfs.h>
8#include <unistd.h>
9#include <fcntl.h>
10#include <errno.h>
11#include <sched.h>
12#include <linux/unistd.h>
13#include <linux/raw.h>
14#include <linux/major.h>
15
16#include "indirect.h"
17
18#define FIO_HAVE_LIBAIO
19#define FIO_HAVE_POSIXAIO
20#define FIO_HAVE_FADVISE
21#define FIO_HAVE_CPU_AFFINITY
22#define FIO_HAVE_DISK_UTIL
23#define FIO_HAVE_SGIO
24#define FIO_HAVE_IOPRIO
25#define FIO_HAVE_SPLICE
26#define FIO_HAVE_IOSCHED_SWITCH
27#define FIO_HAVE_ODIRECT
28#define FIO_HAVE_HUGETLB
29#define FIO_HAVE_RAWBIND
30#define FIO_HAVE_BLKTRACE
31#define FIO_HAVE_STRSEP
32#define FIO_HAVE_FALLOCATE
33#define FIO_HAVE_POSIXAIO_FSYNC
34#define FIO_HAVE_PSHARED_MUTEX
35#define FIO_HAVE_CL_SIZE
36#define FIO_HAVE_CGROUPS
37#define FIO_HAVE_FDATASYNC
38#define FIO_HAVE_FS_STAT
39#define FIO_HAVE_TRIM
40
41#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
42#define FIO_HAVE_SYNC_FILE_RANGE
43#endif
44
45#define OS_MAP_ANON		MAP_ANONYMOUS
46
47#ifndef CLOCK_MONOTONIC
48#define CLOCK_MONOTONIC 1
49#endif
50
51typedef cpu_set_t os_cpu_mask_t;
52
53typedef struct drand48_data os_random_state_t;
54
55/*
56 * we want fadvise64 really, but it's so tangled... later
57 */
58#ifdef FIO_HAVE_FADVISE
59#define fadvise(fd, off, len, advice)	\
60	posix_fadvise((fd), (off_t)(off), (len), (advice))
61#endif
62
63/*
64 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
65 * the affinity helpers to work.
66 */
67#ifndef GLIBC_2_3_2
68#define fio_setaffinity(pid, cpumask)		\
69	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
70#define fio_getaffinity(pid, ptr)	\
71	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
72#else
73#define fio_setaffinity(pid, cpumask)	\
74	sched_setaffinity((pid), &(cpumask))
75#define fio_getaffinity(pid, ptr)	\
76	sched_getaffinity((pid), (ptr))
77#endif
78
79#define fio_cpu_clear(mask, cpu)	CPU_CLR((cpu), (mask))
80#define fio_cpu_set(mask, cpu)		CPU_SET((cpu), (mask))
81
82static inline int fio_cpuset_init(os_cpu_mask_t *mask)
83{
84	CPU_ZERO(mask);
85	return 0;
86}
87
88static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
89{
90	return 0;
91}
92
93#define FIO_MAX_CPUS			CPU_SETSIZE
94
95static inline int ioprio_set(int which, int who, int ioprio)
96{
97	return syscall(__NR_ioprio_set, which, who, ioprio);
98}
99
100/*
101 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
102 * aren't either.
103 */
104#ifndef SPLICE_F_MOVE
105#define SPLICE_F_MOVE	(0x01)	/* move pages instead of copying */
106#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
107				 /* we may still block on the fd we splice */
108				 /* from/to, of course */
109#define SPLICE_F_MORE	(0x04)	/* expect more data */
110#define SPLICE_F_GIFT   (0x08)  /* pages passed in are a gift */
111
112static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
113			 size_t len, unsigned int flags)
114{
115	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
116}
117
118static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
119{
120	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
121}
122
123static inline int vmsplice(int fd, const struct iovec *iov,
124			   unsigned long nr_segs, unsigned int flags)
125{
126	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
127}
128#endif
129
130#define SPLICE_DEF_SIZE	(64*1024)
131
132#ifdef FIO_HAVE_SYSLET
133
134struct syslet_uatom;
135struct async_head_user;
136
137/*
138 * syslet stuff
139 */
140static inline struct syslet_uatom *
141async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
142{
143	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
144}
145
146static inline long
147async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
148	   struct async_head_user *ahu)
149{
150	return syscall(__NR_async_wait, min_wait_events,
151			user_ring_idx, ahu);
152}
153
154static inline long async_thread(void *event, struct async_head_user *ahu)
155{
156	return syscall(__NR_async_thread, event, ahu);
157}
158
159static inline long umem_add(unsigned long *uptr, unsigned long inc)
160{
161	return syscall(__NR_umem_add, uptr, inc);
162}
163#endif /* FIO_HAVE_SYSLET */
164
165enum {
166	IOPRIO_CLASS_NONE,
167	IOPRIO_CLASS_RT,
168	IOPRIO_CLASS_BE,
169	IOPRIO_CLASS_IDLE,
170};
171
172enum {
173	IOPRIO_WHO_PROCESS = 1,
174	IOPRIO_WHO_PGRP,
175	IOPRIO_WHO_USER,
176};
177
178#define IOPRIO_BITS		16
179#define IOPRIO_CLASS_SHIFT	13
180
181#ifndef BLKGETSIZE64
182#define BLKGETSIZE64	_IOR(0x12,114,size_t)
183#endif
184
185#ifndef BLKFLSBUF
186#define BLKFLSBUF	_IO(0x12,97)
187#endif
188
189#ifndef BLKDISCARD
190#define BLKDISCARD	_IO(0x12,119)
191#endif
192
193static inline int blockdev_invalidate_cache(int fd)
194{
195	return ioctl(fd, BLKFLSBUF);
196}
197
198static inline int blockdev_size(int fd, unsigned long long *bytes)
199{
200	if (!ioctl(fd, BLKGETSIZE64, bytes))
201		return 0;
202
203	return errno;
204}
205
206static inline unsigned long long os_phys_mem(void)
207{
208	long pagesize, pages;
209
210	pagesize = sysconf(_SC_PAGESIZE);
211	pages = sysconf(_SC_PHYS_PAGES);
212	if (pages == -1 || pagesize == -1)
213		return 0;
214
215	return (unsigned long long) pages * (unsigned long long) pagesize;
216}
217
218static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
219{
220	srand48_r(seed, rs);
221}
222
223static inline long os_random_long(os_random_state_t *rs)
224{
225	long val;
226
227	lrand48_r(rs, &val);
228	return val;
229}
230
231static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
232{
233	struct raw_config_request rq;
234	int fd;
235
236	if (major(dev) != RAW_MAJOR)
237		return 1;
238
239	/*
240	 * we should be able to find /dev/rawctl or /dev/raw/rawctl
241	 */
242	fd = open("/dev/rawctl", O_RDONLY);
243	if (fd < 0) {
244		fd = open("/dev/raw/rawctl", O_RDONLY);
245		if (fd < 0)
246			return 1;
247	}
248
249	rq.raw_minor = minor(dev);
250	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
251		close(fd);
252		return 1;
253	}
254
255	close(fd);
256	*majdev = rq.block_major;
257	*mindev = rq.block_minor;
258	return 0;
259}
260
261#ifdef O_NOATIME
262#define FIO_O_NOATIME	O_NOATIME
263#else
264#define FIO_O_NOATIME	0
265#endif
266
267#ifdef MADV_REMOVE
268#define FIO_MADV_FREE	MADV_REMOVE
269#endif
270
271#define CACHE_LINE_FILE	\
272	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
273
274static inline int arch_cache_line_size(void)
275{
276	char size[32];
277	int fd, ret;
278
279	fd = open(CACHE_LINE_FILE, O_RDONLY);
280	if (fd < 0)
281		return -1;
282
283	ret = read(fd, size, sizeof(size));
284
285	close(fd);
286
287	if (ret <= 0)
288		return -1;
289	else
290		return atoi(size);
291}
292
293static inline unsigned long long get_fs_size(const char *path)
294{
295	unsigned long long ret;
296	struct statfs s;
297
298	if (statfs(path, &s) < 0)
299		return -1ULL;
300
301	ret = s.f_bsize;
302	ret *= (unsigned long long) s.f_bfree;
303	return ret;
304}
305
306static inline int os_trim(int fd, unsigned long long start,
307			  unsigned long long len)
308{
309	uint64_t range[2];
310
311	range[0] = start;
312	range[1] = len;
313
314	if (!ioctl(fd, BLKDISCARD, range))
315		return 0;
316
317	return errno;
318}
319
320#endif
321