[fio.git] / os / os-linux.h

#ifndef FIO_OS_LINUX_H
#define FIO_OS_LINUX_H

#define	FIO_OS	os_linux

#include <sys/ioctl.h>
#include <sys/uio.h>
#include <sys/syscall.h>
#include <sys/sysmacros.h>
#include <sys/vfs.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include <sched.h>
#include <linux/unistd.h>
#include <linux/raw.h>
#include <linux/major.h>
#include <byteswap.h>

#include "binject.h"
#include "../file.h"

#define FIO_HAVE_CPU_AFFINITY
#define FIO_HAVE_DISK_UTIL
#define FIO_HAVE_SGIO
#define FIO_HAVE_IOPRIO
#define FIO_HAVE_IOSCHED_SWITCH
#define FIO_HAVE_ODIRECT
#define FIO_HAVE_HUGETLB
#define FIO_HAVE_RAWBIND
#define FIO_HAVE_BLKTRACE
#define FIO_HAVE_PSHARED_MUTEX
#define FIO_HAVE_CL_SIZE
#define FIO_HAVE_CGROUPS
#define FIO_HAVE_FS_STAT
#define FIO_HAVE_TRIM
#define FIO_HAVE_BINJECT
#define FIO_HAVE_GETTID
#define FIO_USE_GENERIC_INIT_RANDOM_STATE
#define FIO_HAVE_PWRITEV2

#ifdef MAP_HUGETLB
#define FIO_HAVE_MMAP_HUGE
#endif

#define OS_MAP_ANON		MAP_ANONYMOUS

typedef cpu_set_t os_cpu_mask_t;

typedef struct drand48_data os_random_state_t;

#ifdef CONFIG_3ARG_AFFINITY
#define fio_setaffinity(pid, cpumask)		\
	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
#define fio_getaffinity(pid, ptr)	\
	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
#elif defined(CONFIG_2ARG_AFFINITY)
#define fio_setaffinity(pid, cpumask)	\
	sched_setaffinity((pid), &(cpumask))
#define fio_getaffinity(pid, ptr)	\
	sched_getaffinity((pid), (ptr))
#endif

#define fio_cpu_clear(mask, cpu)	(void) CPU_CLR((cpu), (mask))
#define fio_cpu_set(mask, cpu)		(void) CPU_SET((cpu), (mask))
#define fio_cpu_isset(mask, cpu)	CPU_ISSET((cpu), (mask))
#define fio_cpu_count(mask)		CPU_COUNT((mask))

static inline int fio_cpuset_init(os_cpu_mask_t *mask)
{
	CPU_ZERO(mask);
	return 0;
}

static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
{
	return 0;
}

#define FIO_MAX_CPUS			CPU_SETSIZE

enum {
	IOPRIO_CLASS_NONE,
	IOPRIO_CLASS_RT,
	IOPRIO_CLASS_BE,
	IOPRIO_CLASS_IDLE,
};

enum {
	IOPRIO_WHO_PROCESS = 1,
	IOPRIO_WHO_PGRP,
	IOPRIO_WHO_USER,
};

#define IOPRIO_BITS		16
#define IOPRIO_CLASS_SHIFT	13

static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
{
	/*
	 * If no class is set, assume BE
	 */
	if (!ioprio_class)
		ioprio_class = IOPRIO_CLASS_BE;

	ioprio |= ioprio_class << IOPRIO_CLASS_SHIFT;
	return syscall(__NR_ioprio_set, which, who, ioprio);
}

static inline int gettid(void)
{
	return syscall(__NR_gettid);
}

#define SPLICE_DEF_SIZE	(64*1024)

#ifndef BLKGETSIZE64
#define BLKGETSIZE64	_IOR(0x12,114,size_t)
#endif

#ifndef BLKFLSBUF
#define BLKFLSBUF	_IO(0x12,97)
#endif

#ifndef BLKDISCARD
#define BLKDISCARD	_IO(0x12,119)
#endif

static inline int blockdev_invalidate_cache(struct fio_file *f)
{
	return ioctl(f->fd, BLKFLSBUF);
}

static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
{
	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
		return 0;

	return errno;
}

static inline unsigned long long os_phys_mem(void)
{
	long pagesize, pages;

	pagesize = sysconf(_SC_PAGESIZE);
	pages = sysconf(_SC_PHYS_PAGES);
	if (pages == -1 || pagesize == -1)
		return 0;

	return (unsigned long long) pages * (unsigned long long) pagesize;
}

static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
{
	srand48_r(seed, rs);
}

static inline long os_random_long(os_random_state_t *rs)
{
	long val;

	lrand48_r(rs, &val);
	return val;
}

static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
{
	struct raw_config_request rq;
	int fd;

	if (major(dev) != RAW_MAJOR)
		return 1;

	/*
	 * we should be able to find /dev/rawctl or /dev/raw/rawctl
	 */
	fd = open("/dev/rawctl", O_RDONLY);
	if (fd < 0) {
		fd = open("/dev/raw/rawctl", O_RDONLY);
		if (fd < 0)
			return 1;
	}

	rq.raw_minor = minor(dev);
	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
		close(fd);
		return 1;
	}

	close(fd);
	*majdev = rq.block_major;
	*mindev = rq.block_minor;
	return 0;
}

#ifdef O_NOATIME
#define FIO_O_NOATIME	O_NOATIME
#else
#define FIO_O_NOATIME	0
#endif

#ifdef O_ATOMIC
#define OS_O_ATOMIC	O_ATOMIC
#else
#define OS_O_ATOMIC	040000000
#endif

#ifdef MADV_REMOVE
#define FIO_MADV_FREE	MADV_REMOVE
#endif

#if defined(__builtin_bswap16)
#define fio_swap16(x)	__builtin_bswap16(x)
#else
#define fio_swap16(x)	__bswap_16(x)
#endif
#if defined(__builtin_bswap32)
#define fio_swap32(x)	__builtin_bswap32(x)
#else
#define fio_swap32(x)	__bswap_32(x)
#endif
#if defined(__builtin_bswap64)
#define fio_swap64(x)	__builtin_bswap64(x)
#else
#define fio_swap64(x)	__bswap_64(x)
#endif

#define CACHE_LINE_FILE	\
	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"

static inline int arch_cache_line_size(void)
{
	char size[32];
	int fd, ret;

	fd = open(CACHE_LINE_FILE, O_RDONLY);
	if (fd < 0)
		return -1;

	ret = read(fd, size, sizeof(size));

	close(fd);

	if (ret <= 0)
		return -1;
	else
		return atoi(size);
}

static inline unsigned long long get_fs_free_size(const char *path)
{
	unsigned long long ret;
	struct statfs s;

	if (statfs(path, &s) < 0)
		return -1ULL;

	ret = s.f_bsize;
	ret *= (unsigned long long) s.f_bfree;
	return ret;
}

static inline int os_trim(int fd, unsigned long long start,
			  unsigned long long len)
{
	uint64_t range[2];

	range[0] = start;
	range[1] = len;

	if (!ioctl(fd, BLKDISCARD, range))
		return 0;

	return errno;
}

#ifdef CONFIG_SCHED_IDLE
static inline int fio_set_sched_idle(void)
{
	struct sched_param p = { .sched_priority = 0, };
	return sched_setscheduler(gettid(), SCHED_IDLE, &p);
}
#endif

#ifndef POSIX_FADV_STREAMID
#define POSIX_FADV_STREAMID	8
#endif

#define FIO_HAVE_STREAMID

#ifndef RWF_HIPRI
#define RWF_HIPRI	0x00000001
#endif
#ifndef RWF_DSYNC
#define RWF_DSYNC	0x00000002
#endif
#ifndef RWF_SYNC
#define RWF_SYNC	0x00000004
#endif

#ifndef CONFIG_PWRITEV2
#ifdef __NR_preadv2
static inline void make_pos_h_l(unsigned long *pos_h, unsigned long *pos_l,
				off_t offset)
{
	*pos_l = offset & 0xffffffff;
	*pos_h = ((uint64_t) offset) >> 32;

}
static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
			      off_t offset, unsigned int flags)
{
	unsigned long pos_l, pos_h;

	make_pos_h_l(&pos_h, &pos_l, offset);
	return syscall(__NR_preadv2, fd, iov, iovcnt, pos_l, pos_h, flags);
}
static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
			       off_t offset, unsigned int flags)
{
	unsigned long pos_l, pos_h;

	make_pos_h_l(&pos_h, &pos_l, offset);
	return syscall(__NR_pwritev2, fd, iov, iovcnt, pos_l, pos_h, flags);
}
#else
static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
			      off_t offset, unsigned int flags)
{
	errno = ENOSYS;
	return -1;
}
static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
			       off_t offset, unsigned int flags)
{
	errno = ENOSYS;
	return -1;
}
#endif /* __NR_preadv2 */
#endif /* CONFIG_PWRITEV2 */

#endif
Commit	Line	Data
ebac4655 JA	1	#ifndef FIO_OS_LINUX_H
	2	#define FIO_OS_LINUX_H
	3
cca84643 JA	4	#define FIO_OS os_linux
cca84643 JA	5
ebac4655	6	#include <sys/ioctl.h>
8756e4d4	7	#include <sys/uio.h>
3c39a379	8	#include <sys/syscall.h>
250e878a	9	#include <sys/sysmacros.h>
2e3bd4c2	10	#include <sys/vfs.h>
39b93568	11	#include <sys/mman.h>
3c39a379	12	#include <unistd.h>
97fbdfa1	13	#include <fcntl.h>
3c2d93ed	14	#include <errno.h>
d09913c2	15	#include <sched.h>
ea421790	16	#include <linux/unistd.h>
07e5b264 JA	17	#include <linux/raw.h>
07e5b264 JA	18	#include <linux/major.h>
df73aab1	19	#include <byteswap.h>
ebac4655	20
79a43187	21	#include "binject.h"
e2e58886	22	#include "../file.h"
f3de88a7	23
ebac4655 JA	24	#define FIO_HAVE_CPU_AFFINITY
	25	#define FIO_HAVE_DISK_UTIL
	26	#define FIO_HAVE_SGIO
ba4f8923	27	#define FIO_HAVE_IOPRIO
22f78b32	28	#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28	29	#define FIO_HAVE_ODIRECT
74b025b0	30	#define FIO_HAVE_HUGETLB
07e5b264	31	#define FIO_HAVE_RAWBIND
5e62c22a	32	#define FIO_HAVE_BLKTRACE
f356d01d	33	#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38	34	#define FIO_HAVE_CL_SIZE
a696fa2a	35	#define FIO_HAVE_CGROUPS
2e3bd4c2	36	#define FIO_HAVE_FS_STAT
a5f3027c	37	#define FIO_HAVE_TRIM
79a43187	38	#define FIO_HAVE_BINJECT
47f767c1	39	#define FIO_HAVE_GETTID
93bcfd20	40	#define FIO_USE_GENERIC_INIT_RANDOM_STATE
6562685f	41	#define FIO_HAVE_PWRITEV2
b6959b55	42
39b93568 JA	43	#ifdef MAP_HUGETLB
	44	#define FIO_HAVE_MMAP_HUGE
	45	#endif
	46
dc873b6f	47	#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 JA	48
ebac4655 JA	49	typedef cpu_set_t os_cpu_mask_t;
4e78e405	50
6dfd46b9	51	typedef struct drand48_data os_random_state_t;
ebac4655	52
67bf9823	53	#ifdef CONFIG_3ARG_AFFINITY
e8462bd8 JA	54	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	55	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655 JA	56	#define fio_getaffinity(pid, ptr) \
ebac4655 JA	57	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
67bf9823	58	#elif defined(CONFIG_2ARG_AFFINITY)
e8462bd8 JA	59	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	60	sched_setaffinity((pid), &(cpumask))
c8f025f8 JA	61	#define fio_getaffinity(pid, ptr) \
	62	sched_getaffinity((pid), (ptr))
	63	#endif
be4ecfdf	64
1f4c0a4f JA	65	#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
1f4c0a4f JA	66	#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
50b5860b	67	#define fio_cpu_isset(mask, cpu) CPU_ISSET((cpu), (mask))
d004a209	68	#define fio_cpu_count(mask) CPU_COUNT((mask))
d2ce18b5 JA	69
	70	static inline int fio_cpuset_init(os_cpu_mask_t *mask)
	71	{
	72	CPU_ZERO(mask);
	73	return 0;
	74	}
	75
	76	static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
	77	{
	78	return 0;
	79	}
6d459ee7 JA	80
6d459ee7 JA	81	#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de	82
28727df7 JA	83	enum {
	84	IOPRIO_CLASS_NONE,
	85	IOPRIO_CLASS_RT,
	86	IOPRIO_CLASS_BE,
	87	IOPRIO_CLASS_IDLE,
	88	};
	89
	90	enum {
	91	IOPRIO_WHO_PROCESS = 1,
	92	IOPRIO_WHO_PGRP,
	93	IOPRIO_WHO_USER,
	94	};
	95
	96	#define IOPRIO_BITS 16
	97	#define IOPRIO_CLASS_SHIFT 13
	98
	99	static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
ebac4655	100	{
28727df7 JA	101	/*
	102	* If no class is set, assume BE
	103	*/
	104	if (!ioprio_class)
	105	ioprio_class = IOPRIO_CLASS_BE;
	106
	107	ioprio \|= ioprio_class << IOPRIO_CLASS_SHIFT;
ebac4655 JA	108	return syscall(__NR_ioprio_set, which, who, ioprio);
	109	}
	110
47f767c1 JA	111	static inline int gettid(void)
	112	{
	113	return syscall(__NR_gettid);
	114	}
	115
3feedc60 JA	116	#define SPLICE_DEF_SIZE (64*1024)
3feedc60 JA	117
ebac4655 JA	118	#ifndef BLKGETSIZE64
	119	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	120	#endif
	121
e5b401d4 JA	122	#ifndef BLKFLSBUF
	123	#define BLKFLSBUF _IO(0x12,97)
	124	#endif
	125
a5f3027c JA	126	#ifndef BLKDISCARD
	127	#define BLKDISCARD _IO(0x12,119)
	128	#endif
	129
e2e58886	130	static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4	131	{
ecc314ba	132	return ioctl(f->fd, BLKFLSBUF);
e5b401d4 JA	133	}
e5b401d4 JA	134
ecc314ba	135	static inline int blockdev_size(struct fio_file f, unsigned long long bytes)
ebac4655	136	{
ecc314ba	137	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655 JA	138	return 0;
	139
	140	return errno;
	141	}
	142
32cd46a0 JA	143	static inline unsigned long long os_phys_mem(void)
	144	{
	145	long pagesize, pages;
	146
	147	pagesize = sysconf(_SC_PAGESIZE);
	148	pages = sysconf(_SC_PHYS_PAGES);
	149	if (pages == -1 \|\| pagesize == -1)
	150	return 0;
	151
	152	return (unsigned long long) pages * (unsigned long long) pagesize;
	153	}
	154
6dfd46b9 JA	155	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	156	{
	157	srand48_r(seed, rs);
	158	}
	159
	160	static inline long os_random_long(os_random_state_t *rs)
	161	{
	162	long val;
	163
	164	lrand48_r(rs, &val);
	165	return val;
	166	}
	167
8cc7afa9	168	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
07e5b264 JA	169	{
	170	struct raw_config_request rq;
	171	int fd;
	172
	173	if (major(dev) != RAW_MAJOR)
8cc7afa9	174	return 1;
07e5b264 JA	175
	176	/*
	177	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	178	*/
	179	fd = open("/dev/rawctl", O_RDONLY);
	180	if (fd < 0) {
	181	fd = open("/dev/raw/rawctl", O_RDONLY);
	182	if (fd < 0)
8cc7afa9	183	return 1;
07e5b264 JA	184	}
	185
	186	rq.raw_minor = minor(dev);
	187	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	188	close(fd);
8cc7afa9	189	return 1;
07e5b264 JA	190	}
	191
	192	close(fd);
	193	*majdev = rq.block_major;
	194	*mindev = rq.block_minor;
8cc7afa9	195	return 0;
07e5b264 JA	196	}
07e5b264 JA	197
5b6f5c66	198	#ifdef O_NOATIME
5921e80c	199	#define FIO_O_NOATIME O_NOATIME
5b6f5c66 JA	200	#else
	201	#define FIO_O_NOATIME 0
	202	#endif
5921e80c	203
d01612f3 CM	204	#ifdef O_ATOMIC
	205	#define OS_O_ATOMIC O_ATOMIC
	206	#else
	207	#define OS_O_ATOMIC 040000000
	208	#endif
	209
a1c58075 JA	210	#ifdef MADV_REMOVE
	211	#define FIO_MADV_FREE MADV_REMOVE
	212	#endif
	213
df73aab1 JA	214	#if defined(__builtin_bswap16)
	215	#define fio_swap16(x) __builtin_bswap16(x)
	216	#else
ff245192	217	#define fio_swap16(x) __bswap_16(x)
df73aab1 JA	218	#endif
	219	#if defined(__builtin_bswap32)
	220	#define fio_swap32(x) __builtin_bswap32(x)
	221	#else
ff245192	222	#define fio_swap32(x) __bswap_32(x)
df73aab1 JA	223	#endif
	224	#if defined(__builtin_bswap64)
	225	#define fio_swap64(x) __builtin_bswap64(x)
	226	#else
ff245192	227	#define fio_swap64(x) __bswap_64(x)
df73aab1	228	#endif
ff245192	229
eb7ccf38 JA	230	#define CACHE_LINE_FILE \
	231	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
	232
	233	static inline int arch_cache_line_size(void)
	234	{
	235	char size[32];
	236	int fd, ret;
	237
	238	fd = open(CACHE_LINE_FILE, O_RDONLY);
	239	if (fd < 0)
	240	return -1;
	241
	242	ret = read(fd, size, sizeof(size));
ec76f15a JA	243
	244	close(fd);
	245
eb7ccf38 JA	246	if (ret <= 0)
eb7ccf38 JA	247	return -1;
ec76f15a JA	248	else
ec76f15a JA	249	return atoi(size);
eb7ccf38 JA	250	}
eb7ccf38 JA	251
c08ad04c	252	static inline unsigned long long get_fs_free_size(const char *path)
2e3bd4c2 JA	253	{
	254	unsigned long long ret;
	255	struct statfs s;
	256
	257	if (statfs(path, &s) < 0)
	258	return -1ULL;
	259
	260	ret = s.f_bsize;
	261	ret *= (unsigned long long) s.f_bfree;
	262	return ret;
	263	}
	264
a5f3027c JA	265	static inline int os_trim(int fd, unsigned long long start,
	266	unsigned long long len)
	267	{
	268	uint64_t range[2];
	269
	270	range[0] = start;
	271	range[1] = len;
	272
	273	if (!ioctl(fd, BLKDISCARD, range))
	274	return 0;
	275
	276	return errno;
	277	}
	278
7e09a9f1	279	#ifdef CONFIG_SCHED_IDLE
f2a2ce0e HL	280	static inline int fio_set_sched_idle(void)
	281	{
	282	struct sched_param p = { .sched_priority = 0, };
	283	return sched_setscheduler(gettid(), SCHED_IDLE, &p);
	284	}
7e09a9f1	285	#endif
f2a2ce0e	286
37659335 JA	287	#ifndef POSIX_FADV_STREAMID
	288	#define POSIX_FADV_STREAMID 8
	289	#endif
	290
	291	#define FIO_HAVE_STREAMID
	292
6562685f JA	293	#ifndef RWF_HIPRI
	294	#define RWF_HIPRI 0x00000001
	295	#endif
	296	#ifndef RWF_DSYNC
	297	#define RWF_DSYNC 0x00000002
	298	#endif
	299	#ifndef RWF_SYNC
	300	#define RWF_SYNC 0x00000004
	301	#endif
	302
	303	#ifndef CONFIG_PWRITEV2
	304	#ifdef __NR_preadv2
	305	static inline void make_pos_h_l(unsigned long pos_h, unsigned long pos_l,
	306	off_t offset)
	307	{
	308	*pos_l = offset & 0xffffffff;
	309	*pos_h = ((uint64_t) offset) >> 32;
	310
	311	}
	312	static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
	313	off_t offset, unsigned int flags)
	314	{
	315	unsigned long pos_l, pos_h;
	316
	317	make_pos_h_l(&pos_h, &pos_l, offset);
	318	return syscall(__NR_preadv2, fd, iov, iovcnt, pos_l, pos_h, flags);
	319	}
	320	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	321	off_t offset, unsigned int flags)
	322	{
	323	unsigned long pos_l, pos_h;
	324
	325	make_pos_h_l(&pos_h, &pos_l, offset);
	326	return syscall(__NR_pwritev2, fd, iov, iovcnt, pos_l, pos_h, flags);
	327	}
	328	#else
	329	static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
	330	off_t offset, unsigned int flags)
	331	{
	332	errno = ENOSYS;
	333	return -1;
	334	}
	335	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	336	off_t offset, unsigned int flags)
	337	{
	338	errno = ENOSYS;
	339	return -1;
	340	}
	341	#endif /* __NR_preadv2 */
	342	#endif /* CONFIG_PWRITEV2 */
	343
ebac4655	344	#endif