[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 "./os-linux-syscall.h"
#include "binject.h"
#include "../file.h"

#ifndef __has_builtin         // Optional of course.
  #define __has_builtin(x) 0  // Compatibility with non-clang compilers.
#endif

#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_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
#define FIO_HAVE_SHM_ATTACH_REMOVED

#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

/* Check for GCC or Clang byte swap intrinsics */
#if (__has_builtin(__builtin_bswap16) && __has_builtin(__builtin_bswap32) \
     && __has_builtin(__builtin_bswap64)) || (__GNUC__ > 4 \
     || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) /* fio_swapN */
#define fio_swap16(x)	__builtin_bswap16(x)
#define fio_swap32(x)	__builtin_bswap32(x)
#define fio_swap64(x)	__builtin_bswap64(x)
#else
#include <byteswap.h>
#define fio_swap16(x)	bswap_16(x)
#define fio_swap32(x)	bswap_32(x)
#define fio_swap64(x)	bswap_64(x)
#endif /* fio_swapN */

#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)
{
#if BITS_PER_LONG == 64
	*pos_l = offset;
	*pos_h = 0;
#else
	*pos_l = offset & 0xffffffff;
	*pos_h = ((uint64_t) offset) >> 32;
#endif
}
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 */

static inline int shm_attach_to_open_removed(void)
{
	return 1;
}

#endif
Commit	Line	Data
	1	#ifndef FIO_OS_LINUX_H
	2	#define FIO_OS_LINUX_H
	3
	4	#define FIO_OS os_linux
	5
	6	#include <sys/ioctl.h>
	7	#include <sys/uio.h>
	8	#include <sys/syscall.h>
	9	#include <sys/sysmacros.h>
	10	#include <sys/vfs.h>
	11	#include <sys/mman.h>
	12	#include <unistd.h>
	13	#include <fcntl.h>
	14	#include <errno.h>
	15	#include <sched.h>
	16	#include <linux/unistd.h>
	17	#include <linux/raw.h>
	18	#include <linux/major.h>
	19
	20	#include "./os-linux-syscall.h"
	21	#include "binject.h"
	22	#include "../file.h"
	23
	24	#ifndef __has_builtin // Optional of course.
	25	#define __has_builtin(x) 0 // Compatibility with non-clang compilers.
	26	#endif
	27
	28	#define FIO_HAVE_CPU_AFFINITY
	29	#define FIO_HAVE_DISK_UTIL
	30	#define FIO_HAVE_SGIO
	31	#define FIO_HAVE_IOPRIO
	32	#define FIO_HAVE_IOPRIO_CLASS
	33	#define FIO_HAVE_IOSCHED_SWITCH
	34	#define FIO_HAVE_ODIRECT
	35	#define FIO_HAVE_HUGETLB
	36	#define FIO_HAVE_RAWBIND
	37	#define FIO_HAVE_BLKTRACE
	38	#define FIO_HAVE_CL_SIZE
	39	#define FIO_HAVE_CGROUPS
	40	#define FIO_HAVE_FS_STAT
	41	#define FIO_HAVE_TRIM
	42	#define FIO_HAVE_BINJECT
	43	#define FIO_HAVE_GETTID
	44	#define FIO_USE_GENERIC_INIT_RANDOM_STATE
	45	#define FIO_HAVE_PWRITEV2
	46	#define FIO_HAVE_SHM_ATTACH_REMOVED
	47
	48	#ifdef MAP_HUGETLB
	49	#define FIO_HAVE_MMAP_HUGE
	50	#endif
	51
	52	#define OS_MAP_ANON MAP_ANONYMOUS
	53
	54	typedef cpu_set_t os_cpu_mask_t;
	55
	56	typedef struct drand48_data os_random_state_t;
	57
	58	#ifdef CONFIG_3ARG_AFFINITY
	59	#define fio_setaffinity(pid, cpumask) \
	60	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
	61	#define fio_getaffinity(pid, ptr) \
	62	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
	63	#elif defined(CONFIG_2ARG_AFFINITY)
	64	#define fio_setaffinity(pid, cpumask) \
	65	sched_setaffinity((pid), &(cpumask))
	66	#define fio_getaffinity(pid, ptr) \
	67	sched_getaffinity((pid), (ptr))
	68	#endif
	69
	70	#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
	71	#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
	72	#define fio_cpu_isset(mask, cpu) CPU_ISSET((cpu), (mask))
	73	#define fio_cpu_count(mask) CPU_COUNT((mask))
	74
	75	static inline int fio_cpuset_init(os_cpu_mask_t *mask)
	76	{
	77	CPU_ZERO(mask);
	78	return 0;
	79	}
	80
	81	static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
	82	{
	83	return 0;
	84	}
	85
	86	#define FIO_MAX_CPUS CPU_SETSIZE
	87
	88	enum {
	89	IOPRIO_CLASS_NONE,
	90	IOPRIO_CLASS_RT,
	91	IOPRIO_CLASS_BE,
	92	IOPRIO_CLASS_IDLE,
	93	};
	94
	95	enum {
	96	IOPRIO_WHO_PROCESS = 1,
	97	IOPRIO_WHO_PGRP,
	98	IOPRIO_WHO_USER,
	99	};
	100
	101	#define IOPRIO_BITS 16
	102	#define IOPRIO_CLASS_SHIFT 13
	103
	104	#define IOPRIO_MIN_PRIO 0 /* highest priority */
	105	#define IOPRIO_MAX_PRIO 7 /* lowest priority */
	106
	107	#define IOPRIO_MIN_PRIO_CLASS 0
	108	#define IOPRIO_MAX_PRIO_CLASS 3
	109
	110	static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
	111	{
	112	/*
	113	* If no class is set, assume BE
	114	*/
	115	if (!ioprio_class)
	116	ioprio_class = IOPRIO_CLASS_BE;
	117
	118	ioprio \|= ioprio_class << IOPRIO_CLASS_SHIFT;
	119	return syscall(__NR_ioprio_set, which, who, ioprio);
	120	}
	121
	122	static inline int gettid(void)
	123	{
	124	return syscall(__NR_gettid);
	125	}
	126
	127	#define SPLICE_DEF_SIZE (64*1024)
	128
	129	#ifndef BLKGETSIZE64
	130	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	131	#endif
	132
	133	#ifndef BLKFLSBUF
	134	#define BLKFLSBUF _IO(0x12,97)
	135	#endif
	136
	137	#ifndef BLKDISCARD
	138	#define BLKDISCARD _IO(0x12,119)
	139	#endif
	140
	141	static inline int blockdev_invalidate_cache(struct fio_file *f)
	142	{
	143	return ioctl(f->fd, BLKFLSBUF);
	144	}
	145
	146	static inline int blockdev_size(struct fio_file f, unsigned long long bytes)
	147	{
	148	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
	149	return 0;
	150
	151	return errno;
	152	}
	153
	154	static inline unsigned long long os_phys_mem(void)
	155	{
	156	long pagesize, pages;
	157
	158	pagesize = sysconf(_SC_PAGESIZE);
	159	pages = sysconf(_SC_PHYS_PAGES);
	160	if (pages == -1 \|\| pagesize == -1)
	161	return 0;
	162
	163	return (unsigned long long) pages * (unsigned long long) pagesize;
	164	}
	165
	166	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	167	{
	168	srand48_r(seed, rs);
	169	}
	170
	171	static inline long os_random_long(os_random_state_t *rs)
	172	{
	173	long val;
	174
	175	lrand48_r(rs, &val);
	176	return val;
	177	}
	178
	179	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
	180	{
	181	struct raw_config_request rq;
	182	int fd;
	183
	184	if (major(dev) != RAW_MAJOR)
	185	return 1;
	186
	187	/*
	188	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	189	*/
	190	fd = open("/dev/rawctl", O_RDONLY);
	191	if (fd < 0) {
	192	fd = open("/dev/raw/rawctl", O_RDONLY);
	193	if (fd < 0)
	194	return 1;
	195	}
	196
	197	rq.raw_minor = minor(dev);
	198	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	199	close(fd);
	200	return 1;
	201	}
	202
	203	close(fd);
	204	*majdev = rq.block_major;
	205	*mindev = rq.block_minor;
	206	return 0;
	207	}
	208
	209	#ifdef O_NOATIME
	210	#define FIO_O_NOATIME O_NOATIME
	211	#else
	212	#define FIO_O_NOATIME 0
	213	#endif
	214
	215	#ifdef O_ATOMIC
	216	#define OS_O_ATOMIC O_ATOMIC
	217	#else
	218	#define OS_O_ATOMIC 040000000
	219	#endif
	220
	221	#ifdef MADV_REMOVE
	222	#define FIO_MADV_FREE MADV_REMOVE
	223	#endif
	224
	225	/* Check for GCC or Clang byte swap intrinsics */
	226	#if (__has_builtin(__builtin_bswap16) && __has_builtin(__builtin_bswap32) \
	227	&& __has_builtin(__builtin_bswap64)) \|\| (__GNUC__ > 4 \
	228	\|\| (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) /* fio_swapN */
	229	#define fio_swap16(x) __builtin_bswap16(x)
	230	#define fio_swap32(x) __builtin_bswap32(x)
	231	#define fio_swap64(x) __builtin_bswap64(x)
	232	#else
	233	#include <byteswap.h>
	234	#define fio_swap16(x) bswap_16(x)
	235	#define fio_swap32(x) bswap_32(x)
	236	#define fio_swap64(x) bswap_64(x)
	237	#endif /* fio_swapN */
	238
	239	#define CACHE_LINE_FILE \
	240	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
	241
	242	static inline int arch_cache_line_size(void)
	243	{
	244	char size[32];
	245	int fd, ret;
	246
	247	fd = open(CACHE_LINE_FILE, O_RDONLY);
	248	if (fd < 0)
	249	return -1;
	250
	251	ret = read(fd, size, sizeof(size));
	252
	253	close(fd);
	254
	255	if (ret <= 0)
	256	return -1;
	257	else
	258	return atoi(size);
	259	}
	260
	261	static inline unsigned long long get_fs_free_size(const char *path)
	262	{
	263	unsigned long long ret;
	264	struct statfs s;
	265
	266	if (statfs(path, &s) < 0)
	267	return -1ULL;
	268
	269	ret = s.f_bsize;
	270	ret *= (unsigned long long) s.f_bfree;
	271	return ret;
	272	}
	273
	274	static inline int os_trim(int fd, unsigned long long start,
	275	unsigned long long len)
	276	{
	277	uint64_t range[2];
	278
	279	range[0] = start;
	280	range[1] = len;
	281
	282	if (!ioctl(fd, BLKDISCARD, range))
	283	return 0;
	284
	285	return errno;
	286	}
	287
	288	#ifdef CONFIG_SCHED_IDLE
	289	static inline int fio_set_sched_idle(void)
	290	{
	291	struct sched_param p = { .sched_priority = 0, };
	292	return sched_setscheduler(gettid(), SCHED_IDLE, &p);
	293	}
	294	#endif
	295
	296	#ifndef POSIX_FADV_STREAMID
	297	#define POSIX_FADV_STREAMID 8
	298	#endif
	299
	300	#define FIO_HAVE_STREAMID
	301
	302	#ifndef RWF_HIPRI
	303	#define RWF_HIPRI 0x00000001
	304	#endif
	305	#ifndef RWF_DSYNC
	306	#define RWF_DSYNC 0x00000002
	307	#endif
	308	#ifndef RWF_SYNC
	309	#define RWF_SYNC 0x00000004
	310	#endif
	311
	312	#ifndef CONFIG_PWRITEV2
	313	#ifdef __NR_preadv2
	314	static inline void make_pos_h_l(unsigned long pos_h, unsigned long pos_l,
	315	off_t offset)
	316	{
	317	#if BITS_PER_LONG == 64
	318	*pos_l = offset;
	319	*pos_h = 0;
	320	#else
	321	*pos_l = offset & 0xffffffff;
	322	*pos_h = ((uint64_t) offset) >> 32;
	323	#endif
	324	}
	325	static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
	326	off_t offset, unsigned int flags)
	327	{
	328	unsigned long pos_l, pos_h;
	329
	330	make_pos_h_l(&pos_h, &pos_l, offset);
	331	return syscall(__NR_preadv2, fd, iov, iovcnt, pos_l, pos_h, flags);
	332	}
	333	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	334	off_t offset, unsigned int flags)
	335	{
	336	unsigned long pos_l, pos_h;
	337
	338	make_pos_h_l(&pos_h, &pos_l, offset);
	339	return syscall(__NR_pwritev2, fd, iov, iovcnt, pos_l, pos_h, flags);
	340	}
	341	#else
	342	static inline ssize_t preadv2(int fd, const struct iovec *iov, int iovcnt,
	343	off_t offset, unsigned int flags)
	344	{
	345	errno = ENOSYS;
	346	return -1;
	347	}
	348	static inline ssize_t pwritev2(int fd, const struct iovec *iov, int iovcnt,
	349	off_t offset, unsigned int flags)
	350	{
	351	errno = ENOSYS;
	352	return -1;
	353	}
	354	#endif /* __NR_preadv2 */
	355	#endif /* CONFIG_PWRITEV2 */
	356
	357	static inline int shm_attach_to_open_removed(void)
	358	{
	359	return 1;
	360	}
	361
	362	#endif