[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 "./os-linux-syscall.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_IOPRIO_CLASS
#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

#define IOPRIO_MIN_PRIO		0	/* highest priority */
#define IOPRIO_MAX_PRIO		7	/* lowest priority */

#define IOPRIO_MIN_PRIO_CLASS	0
#define IOPRIO_MAX_PRIO_CLASS	3

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
1c764dbe	21	#include "./os-linux-syscall.h"
79a43187	22	#include "binject.h"
e2e58886	23	#include "../file.h"
f3de88a7	24
ebac4655 JA	25	#define FIO_HAVE_CPU_AFFINITY
	26	#define FIO_HAVE_DISK_UTIL
	27	#define FIO_HAVE_SGIO
ba4f8923	28	#define FIO_HAVE_IOPRIO
32ef447a	29	#define FIO_HAVE_IOPRIO_CLASS
22f78b32	30	#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28	31	#define FIO_HAVE_ODIRECT
74b025b0	32	#define FIO_HAVE_HUGETLB
07e5b264	33	#define FIO_HAVE_RAWBIND
5e62c22a	34	#define FIO_HAVE_BLKTRACE
f356d01d	35	#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38	36	#define FIO_HAVE_CL_SIZE
a696fa2a	37	#define FIO_HAVE_CGROUPS
2e3bd4c2	38	#define FIO_HAVE_FS_STAT
a5f3027c	39	#define FIO_HAVE_TRIM
79a43187	40	#define FIO_HAVE_BINJECT
47f767c1	41	#define FIO_HAVE_GETTID
93bcfd20	42	#define FIO_USE_GENERIC_INIT_RANDOM_STATE
6562685f	43	#define FIO_HAVE_PWRITEV2
b6959b55	44
39b93568 JA	45	#ifdef MAP_HUGETLB
	46	#define FIO_HAVE_MMAP_HUGE
	47	#endif
	48
dc873b6f	49	#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 JA	50
ebac4655 JA	51	typedef cpu_set_t os_cpu_mask_t;
4e78e405	52
6dfd46b9	53	typedef struct drand48_data os_random_state_t;
ebac4655	54
67bf9823	55	#ifdef CONFIG_3ARG_AFFINITY
e8462bd8 JA	56	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	57	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655 JA	58	#define fio_getaffinity(pid, ptr) \
ebac4655 JA	59	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
67bf9823	60	#elif defined(CONFIG_2ARG_AFFINITY)
e8462bd8 JA	61	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	62	sched_setaffinity((pid), &(cpumask))
c8f025f8 JA	63	#define fio_getaffinity(pid, ptr) \
	64	sched_getaffinity((pid), (ptr))
	65	#endif
be4ecfdf	66
1f4c0a4f JA	67	#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
1f4c0a4f JA	68	#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
50b5860b	69	#define fio_cpu_isset(mask, cpu) CPU_ISSET((cpu), (mask))
d004a209	70	#define fio_cpu_count(mask) CPU_COUNT((mask))
d2ce18b5 JA	71
	72	static inline int fio_cpuset_init(os_cpu_mask_t *mask)
	73	{
	74	CPU_ZERO(mask);
	75	return 0;
	76	}
	77
	78	static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
	79	{
	80	return 0;
	81	}
6d459ee7 JA	82
6d459ee7 JA	83	#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de	84
28727df7 JA	85	enum {
	86	IOPRIO_CLASS_NONE,
	87	IOPRIO_CLASS_RT,
	88	IOPRIO_CLASS_BE,
	89	IOPRIO_CLASS_IDLE,
	90	};
	91
	92	enum {
	93	IOPRIO_WHO_PROCESS = 1,
	94	IOPRIO_WHO_PGRP,
	95	IOPRIO_WHO_USER,
	96	};
	97
	98	#define IOPRIO_BITS 16
	99	#define IOPRIO_CLASS_SHIFT 13
	100
1767bd34 TK	101	#define IOPRIO_MIN_PRIO 0 /* highest priority */
	102	#define IOPRIO_MAX_PRIO 7 /* lowest priority */
	103
	104	#define IOPRIO_MIN_PRIO_CLASS 0
	105	#define IOPRIO_MAX_PRIO_CLASS 3
	106
28727df7	107	static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
ebac4655	108	{
28727df7 JA	109	/*
	110	* If no class is set, assume BE
	111	*/
	112	if (!ioprio_class)
	113	ioprio_class = IOPRIO_CLASS_BE;
	114
	115	ioprio \|= ioprio_class << IOPRIO_CLASS_SHIFT;
ebac4655 JA	116	return syscall(__NR_ioprio_set, which, who, ioprio);
	117	}
	118
47f767c1 JA	119	static inline int gettid(void)
	120	{
	121	return syscall(__NR_gettid);
	122	}
	123
3feedc60 JA	124	#define SPLICE_DEF_SIZE (64*1024)
3feedc60 JA	125
ebac4655 JA	126	#ifndef BLKGETSIZE64
	127	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	128	#endif
	129
e5b401d4 JA	130	#ifndef BLKFLSBUF
	131	#define BLKFLSBUF _IO(0x12,97)
	132	#endif
	133
a5f3027c JA	134	#ifndef BLKDISCARD
	135	#define BLKDISCARD _IO(0x12,119)
	136	#endif
	137
e2e58886	138	static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4	139	{
ecc314ba	140	return ioctl(f->fd, BLKFLSBUF);
e5b401d4 JA	141	}
e5b401d4 JA	142
ecc314ba	143	static inline int blockdev_size(struct fio_file f, unsigned long long bytes)
ebac4655	144	{
ecc314ba	145	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655 JA	146	return 0;
	147
	148	return errno;
	149	}
	150
32cd46a0 JA	151	static inline unsigned long long os_phys_mem(void)
	152	{
	153	long pagesize, pages;
	154
	155	pagesize = sysconf(_SC_PAGESIZE);
	156	pages = sysconf(_SC_PHYS_PAGES);
	157	if (pages == -1 \|\| pagesize == -1)
	158	return 0;
	159
	160	return (unsigned long long) pages * (unsigned long long) pagesize;
	161	}
	162
6dfd46b9 JA	163	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	164	{
	165	srand48_r(seed, rs);
	166	}
	167
	168	static inline long os_random_long(os_random_state_t *rs)
	169	{
	170	long val;
	171
	172	lrand48_r(rs, &val);
	173	return val;
	174	}
	175
8cc7afa9	176	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
07e5b264 JA	177	{
	178	struct raw_config_request rq;
	179	int fd;
	180
	181	if (major(dev) != RAW_MAJOR)
8cc7afa9	182	return 1;
07e5b264 JA	183
	184	/*
	185	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	186	*/
	187	fd = open("/dev/rawctl", O_RDONLY);
	188	if (fd < 0) {
	189	fd = open("/dev/raw/rawctl", O_RDONLY);
	190	if (fd < 0)
8cc7afa9	191	return 1;
07e5b264 JA	192	}
	193
	194	rq.raw_minor = minor(dev);
	195	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	196	close(fd);
8cc7afa9	197	return 1;
07e5b264 JA	198	}
	199
	200	close(fd);
	201	*majdev = rq.block_major;
	202	*mindev = rq.block_minor;
8cc7afa9	203	return 0;
07e5b264 JA	204	}
07e5b264 JA	205
5b6f5c66	206	#ifdef O_NOATIME
5921e80c	207	#define FIO_O_NOATIME O_NOATIME
5b6f5c66 JA	208	#else
	209	#define FIO_O_NOATIME 0
	210	#endif
5921e80c	211
d01612f3 CM	212	#ifdef O_ATOMIC
	213	#define OS_O_ATOMIC O_ATOMIC
	214	#else
	215	#define OS_O_ATOMIC 040000000
	216	#endif
	217
a1c58075 JA	218	#ifdef MADV_REMOVE
	219	#define FIO_MADV_FREE MADV_REMOVE
	220	#endif
	221
df73aab1 JA	222	#if defined(__builtin_bswap16)
	223	#define fio_swap16(x) __builtin_bswap16(x)
	224	#else
ff245192	225	#define fio_swap16(x) __bswap_16(x)
df73aab1 JA	226	#endif
	227	#if defined(__builtin_bswap32)
	228	#define fio_swap32(x) __builtin_bswap32(x)
	229	#else
ff245192	230	#define fio_swap32(x) __bswap_32(x)
df73aab1 JA	231	#endif
	232	#if defined(__builtin_bswap64)
	233	#define fio_swap64(x) __builtin_bswap64(x)
	234	#else
ff245192	235	#define fio_swap64(x) __bswap_64(x)
df73aab1	236	#endif
ff245192	237
eb7ccf38 JA	238	#define CACHE_LINE_FILE \
	239	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
	240
	241	static inline int arch_cache_line_size(void)
	242	{
	243	char size[32];
	244	int fd, ret;
	245
	246	fd = open(CACHE_LINE_FILE, O_RDONLY);
	247	if (fd < 0)
	248	return -1;
	249
	250	ret = read(fd, size, sizeof(size));
ec76f15a JA	251
	252	close(fd);
	253
eb7ccf38 JA	254	if (ret <= 0)
eb7ccf38 JA	255	return -1;
ec76f15a JA	256	else
ec76f15a JA	257	return atoi(size);
eb7ccf38 JA	258	}
eb7ccf38 JA	259
c08ad04c	260	static inline unsigned long long get_fs_free_size(const char *path)
2e3bd4c2 JA	261	{
	262	unsigned long long ret;
	263	struct statfs s;
	264
	265	if (statfs(path, &s) < 0)
	266	return -1ULL;
	267
	268	ret = s.f_bsize;
	269	ret *= (unsigned long long) s.f_bfree;
	270	return ret;
	271	}
	272
a5f3027c JA	273	static inline int os_trim(int fd, unsigned long long start,
	274	unsigned long long len)
	275	{
	276	uint64_t range[2];
	277
	278	range[0] = start;
	279	range[1] = len;
	280
	281	if (!ioctl(fd, BLKDISCARD, range))
	282	return 0;
	283
	284	return errno;
	285	}
	286
7e09a9f1	287	#ifdef CONFIG_SCHED_IDLE
f2a2ce0e HL	288	static inline int fio_set_sched_idle(void)
	289	{
	290	struct sched_param p = { .sched_priority = 0, };
	291	return sched_setscheduler(gettid(), SCHED_IDLE, &p);
	292	}
7e09a9f1	293	#endif
f2a2ce0e	294
37659335 JA	295	#ifndef POSIX_FADV_STREAMID
	296	#define POSIX_FADV_STREAMID 8
	297	#endif
	298
	299	#define FIO_HAVE_STREAMID
	300
6562685f JA	301	#ifndef RWF_HIPRI
	302	#define RWF_HIPRI 0x00000001
	303	#endif
	304	#ifndef RWF_DSYNC
	305	#define RWF_DSYNC 0x00000002
	306	#endif
	307	#ifndef RWF_SYNC
	308	#define RWF_SYNC 0x00000004
	309	#endif
	310
	311	#ifndef CONFIG_PWRITEV2
	312	#ifdef __NR_preadv2
	313	static inline void make_pos_h_l(unsigned long pos_h, unsigned long pos_l,
	314	off_t offset)
	315	{
	316	*pos_l = offset & 0xffffffff;
	317	*pos_h = ((uint64_t) offset) >> 32;
	318
	319	}
	320	static inline ssize_t preadv2(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_preadv2, fd, iov, iovcnt, pos_l, pos_h, flags);
	327	}
	328	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	329	off_t offset, unsigned int flags)
	330	{
	331	unsigned long pos_l, pos_h;
	332
	333	make_pos_h_l(&pos_h, &pos_l, offset);
	334	return syscall(__NR_pwritev2, fd, iov, iovcnt, pos_l, pos_h, flags);
	335	}
	336	#else
	337	static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
	338	off_t offset, unsigned int flags)
	339	{
	340	errno = ENOSYS;
	341	return -1;
	342	}
	343	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	344	off_t offset, unsigned int flags)
	345	{
	346	errno = ENOSYS;
	347	return -1;
	348	}
	349	#endif /* __NR_preadv2 */
	350	#endif /* CONFIG_PWRITEV2 */
	351
ebac4655	352	#endif