[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/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 <endian.h>

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

#define FIO_HAVE_LIBAIO
#define FIO_HAVE_POSIXAIO
#define FIO_HAVE_FADVISE
#define FIO_HAVE_CPU_AFFINITY
#define FIO_HAVE_DISK_UTIL
#define FIO_HAVE_SGIO
#define FIO_HAVE_IOPRIO
#define FIO_HAVE_SPLICE
#define FIO_HAVE_IOSCHED_SWITCH
#define FIO_HAVE_ODIRECT
#define FIO_HAVE_HUGETLB
#define FIO_HAVE_RAWBIND
#define FIO_HAVE_BLKTRACE
#define FIO_HAVE_STRSEP
#define FIO_HAVE_POSIXAIO_FSYNC
#define FIO_HAVE_PSHARED_MUTEX
#define FIO_HAVE_CL_SIZE
#define FIO_HAVE_CGROUPS
#define FIO_HAVE_FDATASYNC
#define FIO_HAVE_FS_STAT
#define FIO_HAVE_TRIM
#define FIO_HAVE_BINJECT
#define FIO_HAVE_CLOCK_MONOTONIC
#define FIO_HAVE_GETTID
#define FIO_USE_GENERIC_INIT_RANDOM_STATE
#define FIO_HAVE_E4_ENG

#ifdef MAP_HUGETLB
#define FIO_HAVE_MMAP_HUGE
#endif

/*
 * Can only enable this for newer glibcs, or the header and defines are
 * missing
 */
#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
#define FIO_HAVE_FALLOCATE
#endif
#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
#define FIO_HAVE_LINUX_FALLOCATE
#endif

#ifdef FIO_HAVE_LINUX_FALLOCATE
#define FIO_HAVE_FALLOC_ENG
#endif

#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
#define FIO_HAVE_SYNC_FILE_RANGE
#endif

#define OS_MAP_ANON		MAP_ANONYMOUS

#ifndef CLOCK_MONOTONIC
#define CLOCK_MONOTONIC 1
#endif

typedef cpu_set_t os_cpu_mask_t;

typedef struct drand48_data os_random_state_t;

/*
 * we want fadvise64 really, but it's so tangled... later
 */
#ifdef FIO_HAVE_FADVISE
#define fadvise(fd, off, len, advice)	\
	posix_fadvise((fd), (off_t)(off), (len), (advice))
#endif

/*
 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
 * the affinity helpers to work.
 */
#ifndef GLIBC_2_3_2
#define fio_setaffinity(pid, cpumask)		\
	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
#define fio_getaffinity(pid, ptr)	\
	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
#else
#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))

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

static inline int ioprio_set(int which, int who, int ioprio)
{
	return syscall(__NR_ioprio_set, which, who, ioprio);
}

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

/*
 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
 * aren't either.
 */
#ifndef SPLICE_F_MOVE
#define SPLICE_F_MOVE	(0x01)	/* move pages instead of copying */
#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
				 /* we may still block on the fd we splice */
				 /* from/to, of course */
#define SPLICE_F_MORE	(0x04)	/* expect more data */
#define SPLICE_F_GIFT   (0x08)  /* pages passed in are a gift */

static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
			 size_t len, unsigned int flags)
{
	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
}

static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
{
	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
}

static inline int vmsplice(int fd, const struct iovec *iov,
			   unsigned long nr_segs, unsigned int flags)
{
	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
}
#endif

#define SPLICE_DEF_SIZE	(64*1024)

#ifdef FIO_HAVE_SYSLET

struct syslet_uatom;
struct async_head_user;

/*
 * syslet stuff
 */
static inline struct syslet_uatom *
async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
{
	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
}

static inline long
async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
	   struct async_head_user *ahu)
{
	return syscall(__NR_async_wait, min_wait_events,
			user_ring_idx, ahu);
}

static inline long async_thread(void *event, struct async_head_user *ahu)
{
	return syscall(__NR_async_thread, event, ahu);
}

static inline long umem_add(unsigned long *uptr, unsigned long inc)
{
	return syscall(__NR_umem_add, uptr, inc);
}
#endif /* FIO_HAVE_SYSLET */

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

#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 MADV_REMOVE
#define FIO_MADV_FREE	MADV_REMOVE
#endif

#if __BYTE_ORDER == __LITTLE_ENDIAN
#define FIO_LITTLE_ENDIAN
#elif __BYTE_ORDER == __BIG_ENDIAN
#define FIO_BIG_ENDIAN
#else
#error "Unknown endianness"
#endif

#define fio_swap16(x)	__bswap_16(x)
#define fio_swap32(x)	__bswap_32(x)
#define fio_swap64(x)	__bswap_64(x)

#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_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;
}

#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/vfs.h>
	10	#include <sys/mman.h>
	11	#include <unistd.h>
	12	#include <fcntl.h>
	13	#include <errno.h>
	14	#include <sched.h>
	15	#include <linux/unistd.h>
	16	#include <linux/raw.h>
	17	#include <linux/major.h>
	18	#include <endian.h>
	19
	20	#include "indirect.h"
	21	#include "binject.h"
	22	#include "../file.h"
	23
	24	#define FIO_HAVE_LIBAIO
	25	#define FIO_HAVE_POSIXAIO
	26	#define FIO_HAVE_FADVISE
	27	#define FIO_HAVE_CPU_AFFINITY
	28	#define FIO_HAVE_DISK_UTIL
	29	#define FIO_HAVE_SGIO
	30	#define FIO_HAVE_IOPRIO
	31	#define FIO_HAVE_SPLICE
	32	#define FIO_HAVE_IOSCHED_SWITCH
	33	#define FIO_HAVE_ODIRECT
	34	#define FIO_HAVE_HUGETLB
	35	#define FIO_HAVE_RAWBIND
	36	#define FIO_HAVE_BLKTRACE
	37	#define FIO_HAVE_STRSEP
	38	#define FIO_HAVE_POSIXAIO_FSYNC
	39	#define FIO_HAVE_PSHARED_MUTEX
	40	#define FIO_HAVE_CL_SIZE
	41	#define FIO_HAVE_CGROUPS
	42	#define FIO_HAVE_FDATASYNC
	43	#define FIO_HAVE_FS_STAT
	44	#define FIO_HAVE_TRIM
	45	#define FIO_HAVE_BINJECT
	46	#define FIO_HAVE_CLOCK_MONOTONIC
	47	#define FIO_HAVE_GETTID
	48	#define FIO_USE_GENERIC_INIT_RANDOM_STATE
	49	#define FIO_HAVE_E4_ENG
	50
	51	#ifdef MAP_HUGETLB
	52	#define FIO_HAVE_MMAP_HUGE
	53	#endif
	54
	55	/*
	56	* Can only enable this for newer glibcs, or the header and defines are
	57	* missing
	58	*/
	59	#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
	60	#define FIO_HAVE_FALLOCATE
	61	#endif
	62	#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
	63	#define FIO_HAVE_LINUX_FALLOCATE
	64	#endif
	65
	66	#ifdef FIO_HAVE_LINUX_FALLOCATE
	67	#define FIO_HAVE_FALLOC_ENG
	68	#endif
	69
	70	#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
	71	#define FIO_HAVE_SYNC_FILE_RANGE
	72	#endif
	73
	74	#define OS_MAP_ANON MAP_ANONYMOUS
	75
	76	#ifndef CLOCK_MONOTONIC
	77	#define CLOCK_MONOTONIC 1
	78	#endif
	79
	80	typedef cpu_set_t os_cpu_mask_t;
	81
	82	typedef struct drand48_data os_random_state_t;
	83
	84	/*
	85	* we want fadvise64 really, but it's so tangled... later
	86	*/
	87	#ifdef FIO_HAVE_FADVISE
	88	#define fadvise(fd, off, len, advice) \
	89	posix_fadvise((fd), (off_t)(off), (len), (advice))
	90	#endif
	91
	92	/*
	93	* If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
	94	* the affinity helpers to work.
	95	*/
	96	#ifndef GLIBC_2_3_2
	97	#define fio_setaffinity(pid, cpumask) \
	98	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
	99	#define fio_getaffinity(pid, ptr) \
	100	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
	101	#else
	102	#define fio_setaffinity(pid, cpumask) \
	103	sched_setaffinity((pid), &(cpumask))
	104	#define fio_getaffinity(pid, ptr) \
	105	sched_getaffinity((pid), (ptr))
	106	#endif
	107
	108	#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
	109	#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
	110
	111	static inline int fio_cpuset_init(os_cpu_mask_t *mask)
	112	{
	113	CPU_ZERO(mask);
	114	return 0;
	115	}
	116
	117	static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
	118	{
	119	return 0;
	120	}
	121
	122	#define FIO_MAX_CPUS CPU_SETSIZE
	123
	124	static inline int ioprio_set(int which, int who, int ioprio)
	125	{
	126	return syscall(__NR_ioprio_set, which, who, ioprio);
	127	}
	128
	129	static inline int gettid(void)
	130	{
	131	return syscall(__NR_gettid);
	132	}
	133
	134	/*
	135	* Just check for SPLICE_F_MOVE, if that isn't there, assume the others
	136	* aren't either.
	137	*/
	138	#ifndef SPLICE_F_MOVE
	139	#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
	140	#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
	141	/* we may still block on the fd we splice */
	142	/* from/to, of course */
	143	#define SPLICE_F_MORE (0x04) /* expect more data */
	144	#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
	145
	146	static inline int splice(int fdin, loff_t off_in, int fdout, loff_t off_out,
	147	size_t len, unsigned int flags)
	148	{
	149	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
	150	}
	151
	152	static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
	153	{
	154	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
	155	}
	156
	157	static inline int vmsplice(int fd, const struct iovec *iov,
	158	unsigned long nr_segs, unsigned int flags)
	159	{
	160	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
	161	}
	162	#endif
	163
	164	#define SPLICE_DEF_SIZE (64*1024)
	165
	166	#ifdef FIO_HAVE_SYSLET
	167
	168	struct syslet_uatom;
	169	struct async_head_user;
	170
	171	/*
	172	* syslet stuff
	173	*/
	174	static inline struct syslet_uatom *
	175	async_exec(struct syslet_uatom atom, struct async_head_user ahu)
	176	{
	177	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
	178	}
	179
	180	static inline long
	181	async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
	182	struct async_head_user *ahu)
	183	{
	184	return syscall(__NR_async_wait, min_wait_events,
	185	user_ring_idx, ahu);
	186	}
	187
	188	static inline long async_thread(void event, struct async_head_user ahu)
	189	{
	190	return syscall(__NR_async_thread, event, ahu);
	191	}
	192
	193	static inline long umem_add(unsigned long *uptr, unsigned long inc)
	194	{
	195	return syscall(__NR_umem_add, uptr, inc);
	196	}
	197	#endif /* FIO_HAVE_SYSLET */
	198
	199	enum {
	200	IOPRIO_CLASS_NONE,
	201	IOPRIO_CLASS_RT,
	202	IOPRIO_CLASS_BE,
	203	IOPRIO_CLASS_IDLE,
	204	};
	205
	206	enum {
	207	IOPRIO_WHO_PROCESS = 1,
	208	IOPRIO_WHO_PGRP,
	209	IOPRIO_WHO_USER,
	210	};
	211
	212	#define IOPRIO_BITS 16
	213	#define IOPRIO_CLASS_SHIFT 13
	214
	215	#ifndef BLKGETSIZE64
	216	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	217	#endif
	218
	219	#ifndef BLKFLSBUF
	220	#define BLKFLSBUF _IO(0x12,97)
	221	#endif
	222
	223	#ifndef BLKDISCARD
	224	#define BLKDISCARD _IO(0x12,119)
	225	#endif
	226
	227	static inline int blockdev_invalidate_cache(struct fio_file *f)
	228	{
	229	return ioctl(f->fd, BLKFLSBUF);
	230	}
	231
	232	static inline int blockdev_size(struct fio_file f, unsigned long long bytes)
	233	{
	234	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
	235	return 0;
	236
	237	return errno;
	238	}
	239
	240	static inline unsigned long long os_phys_mem(void)
	241	{
	242	long pagesize, pages;
	243
	244	pagesize = sysconf(_SC_PAGESIZE);
	245	pages = sysconf(_SC_PHYS_PAGES);
	246	if (pages == -1 \|\| pagesize == -1)
	247	return 0;
	248
	249	return (unsigned long long) pages * (unsigned long long) pagesize;
	250	}
	251
	252	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	253	{
	254	srand48_r(seed, rs);
	255	}
	256
	257	static inline long os_random_long(os_random_state_t *rs)
	258	{
	259	long val;
	260
	261	lrand48_r(rs, &val);
	262	return val;
	263	}
	264
	265	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
	266	{
	267	struct raw_config_request rq;
	268	int fd;
	269
	270	if (major(dev) != RAW_MAJOR)
	271	return 1;
	272
	273	/*
	274	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	275	*/
	276	fd = open("/dev/rawctl", O_RDONLY);
	277	if (fd < 0) {
	278	fd = open("/dev/raw/rawctl", O_RDONLY);
	279	if (fd < 0)
	280	return 1;
	281	}
	282
	283	rq.raw_minor = minor(dev);
	284	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	285	close(fd);
	286	return 1;
	287	}
	288
	289	close(fd);
	290	*majdev = rq.block_major;
	291	*mindev = rq.block_minor;
	292	return 0;
	293	}
	294
	295	#ifdef O_NOATIME
	296	#define FIO_O_NOATIME O_NOATIME
	297	#else
	298	#define FIO_O_NOATIME 0
	299	#endif
	300
	301	#ifdef MADV_REMOVE
	302	#define FIO_MADV_FREE MADV_REMOVE
	303	#endif
	304
	305	#if __BYTE_ORDER == __LITTLE_ENDIAN
	306	#define FIO_LITTLE_ENDIAN
	307	#elif __BYTE_ORDER == __BIG_ENDIAN
	308	#define FIO_BIG_ENDIAN
	309	#else
	310	#error "Unknown endianness"
	311	#endif
	312
	313	#define fio_swap16(x) __bswap_16(x)
	314	#define fio_swap32(x) __bswap_32(x)
	315	#define fio_swap64(x) __bswap_64(x)
	316
	317	#define CACHE_LINE_FILE \
	318	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
	319
	320	static inline int arch_cache_line_size(void)
	321	{
	322	char size[32];
	323	int fd, ret;
	324
	325	fd = open(CACHE_LINE_FILE, O_RDONLY);
	326	if (fd < 0)
	327	return -1;
	328
	329	ret = read(fd, size, sizeof(size));
	330
	331	close(fd);
	332
	333	if (ret <= 0)
	334	return -1;
	335	else
	336	return atoi(size);
	337	}
	338
	339	static inline unsigned long long get_fs_size(const char *path)
	340	{
	341	unsigned long long ret;
	342	struct statfs s;
	343
	344	if (statfs(path, &s) < 0)
	345	return -1ULL;
	346
	347	ret = s.f_bsize;
	348	ret *= (unsigned long long) s.f_bfree;
	349	return ret;
	350	}
	351
	352	static inline int os_trim(int fd, unsigned long long start,
	353	unsigned long long len)
	354	{
	355	uint64_t range[2];
	356
	357	range[0] = start;
	358	range[1] = len;
	359
	360	if (!ioctl(fd, BLKDISCARD, range))
	361	return 0;
	362
	363	return errno;
	364	}
	365
	366	#endif