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

#ifndef FIO_OS_LINUX_H
#define FIO_OS_LINUX_H

#include <sys/ioctl.h>
#include <sys/uio.h>
#include <sys/syscall.h>
#include <unistd.h>
#include <fcntl.h>
#include <linux/unistd.h>
#include <linux/raw.h>
#include <linux/major.h>

#include "indirect.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_FALLOCATE
#define FIO_HAVE_POSIXAIO_FSYNC

#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(td)		\
	sched_setaffinity((td)->pid, sizeof((td)->o.cpumask), &(td)->o.cpumask)
#define fio_getaffinity(pid, ptr)	\
	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
#else
#define fio_setaffinity(td)		\
	sched_setaffinity((td)->pid, &(td)->o.cpumask)
#define fio_getaffinity(pid, ptr)	\
	sched_getaffinity((pid), (ptr))
#endif

#define fio_cpu_clear(mask, cpu)	CPU_CLR((cpu), (mask))
#define fio_cpu_set(mask, cpu)		CPU_SET((cpu), (mask))
#define fio_cpuset_init(td)		CPU_ZERO(&(td)->o.cpumask)
#define fio_cpuset_exit(td)		do { } while (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);
}

/*
 * 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

static inline int blockdev_invalidate_cache(int fd)
{
	return ioctl(fd, BLKFLSBUF);
}

static inline int blockdev_size(int fd, unsigned long long *bytes)
{
	if (!ioctl(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

#endif
Commit	Line	Data
ebac4655 JA	1	#ifndef FIO_OS_LINUX_H
	2	#define FIO_OS_LINUX_H
	3
	4	#include <sys/ioctl.h>
8756e4d4	5	#include <sys/uio.h>
3c39a379 JA	6	#include <sys/syscall.h>
3c39a379 JA	7	#include <unistd.h>
97fbdfa1	8	#include <fcntl.h>
ea421790	9	#include <linux/unistd.h>
07e5b264 JA	10	#include <linux/raw.h>
07e5b264 JA	11	#include <linux/major.h>
ebac4655	12
f3de88a7 JA	13	#include "indirect.h"
f3de88a7 JA	14
ebac4655 JA	15	#define FIO_HAVE_LIBAIO
	16	#define FIO_HAVE_POSIXAIO
	17	#define FIO_HAVE_FADVISE
	18	#define FIO_HAVE_CPU_AFFINITY
	19	#define FIO_HAVE_DISK_UTIL
	20	#define FIO_HAVE_SGIO
ba4f8923	21	#define FIO_HAVE_IOPRIO
8756e4d4	22	#define FIO_HAVE_SPLICE
22f78b32	23	#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28	24	#define FIO_HAVE_ODIRECT
74b025b0	25	#define FIO_HAVE_HUGETLB
07e5b264	26	#define FIO_HAVE_RAWBIND
5e62c22a	27	#define FIO_HAVE_BLKTRACE
5921e80c	28	#define FIO_HAVE_STRSEP
fffca02d	29	#define FIO_HAVE_FALLOCATE
207cb0f0	30	#define FIO_HAVE_POSIXAIO_FSYNC
b6959b55	31
dc873b6f	32	#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655	33
4d8947de SL	34	#ifndef CLOCK_MONOTONIC
	35	#define CLOCK_MONOTONIC 1
	36	#endif
	37
ebac4655	38	typedef cpu_set_t os_cpu_mask_t;
4e78e405	39
6dfd46b9	40	typedef struct drand48_data os_random_state_t;
ebac4655 JA	41
	42	/*
	43	* we want fadvise64 really, but it's so tangled... later
	44	*/
4d8947de	45	#ifdef FIO_HAVE_FADVISE
ebac4655 JA	46	#define fadvise(fd, off, len, advice) \
ebac4655 JA	47	posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de	48	#endif
ebac4655	49
c8f025f8 JA	50	/*
	51	* If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
	52	* the affinity helpers to work.
	53	*/
	54	#ifndef GLIBC_2_3_2
ebac4655	55	#define fio_setaffinity(td) \
2dc1bbeb	56	sched_setaffinity((td)->pid, sizeof((td)->o.cpumask), &(td)->o.cpumask)
ebac4655 JA	57	#define fio_getaffinity(pid, ptr) \
ebac4655 JA	58	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8 JA	59	#else
	60	#define fio_setaffinity(td) \
	61	sched_setaffinity((td)->pid, &(td)->o.cpumask)
	62	#define fio_getaffinity(pid, ptr) \
	63	sched_getaffinity((pid), (ptr))
	64	#endif
be4ecfdf JA	65
be4ecfdf JA	66	#define fio_cpu_clear(mask, cpu) CPU_CLR((cpu), (mask))
6d459ee7 JA	67	#define fio_cpu_set(mask, cpu) CPU_SET((cpu), (mask))
6d459ee7 JA	68	#define fio_cpuset_init(td) CPU_ZERO(&(td)->o.cpumask)
4e78e405	69	#define fio_cpuset_exit(td) do { } while (0)
6d459ee7 JA	70
6d459ee7 JA	71	#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de	72
ebac4655 JA	73	static inline int ioprio_set(int which, int who, int ioprio)
	74	{
	75	return syscall(__NR_ioprio_set, which, who, ioprio);
	76	}
	77
97fbdfa1 JA	78	/*
	79	* Just check for SPLICE_F_MOVE, if that isn't there, assume the others
	80	* aren't either.
	81	*/
	82	#ifndef SPLICE_F_MOVE
	83	#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
	84	#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
	85	/* we may still block on the fd we splice */
	86	/* from/to, of course */
	87	#define SPLICE_F_MORE (0x04) /* expect more data */
	88	#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4 JA	89
8756e4d4 JA	90	static inline int splice(int fdin, loff_t off_in, int fdout, loff_t off_out,
495ee9b6	91	size_t len, unsigned int flags)
8756e4d4	92	{
97fbdfa1	93	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4 JA	94	}
	95
	96	static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
	97	{
97fbdfa1	98	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4 JA	99	}
	100
	101	static inline int vmsplice(int fd, const struct iovec *iov,
	102	unsigned long nr_segs, unsigned int flags)
	103	{
97fbdfa1	104	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4	105	}
97fbdfa1	106	#endif
8756e4d4	107
3feedc60 JA	108	#define SPLICE_DEF_SIZE (64*1024)
3feedc60 JA	109
b6959b55	110	#ifdef FIO_HAVE_SYSLET
bf0dc8fa IM	111
	112	struct syslet_uatom;
	113	struct async_head_user;
	114
a4f4fdd7 JA	115	/*
	116	* syslet stuff
	117	*/
bf0dc8fa IM	118	static inline struct syslet_uatom *
bf0dc8fa IM	119	async_exec(struct syslet_uatom atom, struct async_head_user ahu)
a4f4fdd7	120	{
2b52511f	121	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7 JA	122	}
a4f4fdd7 JA	123
bf0dc8fa IM	124	static inline long
	125	async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
	126	struct async_head_user *ahu)
a4f4fdd7	127	{
bf0dc8fa IM	128	return syscall(__NR_async_wait, min_wait_events,
bf0dc8fa IM	129	user_ring_idx, ahu);
a4f4fdd7 JA	130	}
a4f4fdd7 JA	131
7756b0d0	132	static inline long async_thread(void event, struct async_head_user ahu)
a4f4fdd7	133	{
7756b0d0	134	return syscall(__NR_async_thread, event, ahu);
a4f4fdd7 JA	135	}
	136
	137	static inline long umem_add(unsigned long *uptr, unsigned long inc)
	138	{
	139	return syscall(__NR_umem_add, uptr, inc);
	140	}
b6959b55	141	#endif /* FIO_HAVE_SYSLET */
a4f4fdd7	142
6cefbe33 JA	143	enum {
	144	IOPRIO_CLASS_NONE,
	145	IOPRIO_CLASS_RT,
	146	IOPRIO_CLASS_BE,
	147	IOPRIO_CLASS_IDLE,
	148	};
	149
ebac4655 JA	150	enum {
	151	IOPRIO_WHO_PROCESS = 1,
	152	IOPRIO_WHO_PGRP,
	153	IOPRIO_WHO_USER,
	154	};
	155
6cefbe33	156	#define IOPRIO_BITS 16
ebac4655 JA	157	#define IOPRIO_CLASS_SHIFT 13
	158
	159	#ifndef BLKGETSIZE64
	160	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	161	#endif
	162
e5b401d4 JA	163	#ifndef BLKFLSBUF
	164	#define BLKFLSBUF _IO(0x12,97)
	165	#endif
	166
	167	static inline int blockdev_invalidate_cache(int fd)
	168	{
a16211bb	169	return ioctl(fd, BLKFLSBUF);
e5b401d4 JA	170	}
e5b401d4 JA	171
9104f874	172	static inline int blockdev_size(int fd, unsigned long long *bytes)
ebac4655 JA	173	{
	174	if (!ioctl(fd, BLKGETSIZE64, bytes))
	175	return 0;
	176
	177	return errno;
	178	}
	179
32cd46a0 JA	180	static inline unsigned long long os_phys_mem(void)
	181	{
	182	long pagesize, pages;
	183
	184	pagesize = sysconf(_SC_PAGESIZE);
	185	pages = sysconf(_SC_PHYS_PAGES);
	186	if (pages == -1 \|\| pagesize == -1)
	187	return 0;
	188
	189	return (unsigned long long) pages * (unsigned long long) pagesize;
	190	}
	191
6dfd46b9 JA	192	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	193	{
	194	srand48_r(seed, rs);
	195	}
	196
	197	static inline long os_random_long(os_random_state_t *rs)
	198	{
	199	long val;
	200
	201	lrand48_r(rs, &val);
	202	return val;
	203	}
	204
8cc7afa9	205	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
07e5b264 JA	206	{
	207	struct raw_config_request rq;
	208	int fd;
	209
	210	if (major(dev) != RAW_MAJOR)
8cc7afa9	211	return 1;
07e5b264 JA	212
	213	/*
	214	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	215	*/
	216	fd = open("/dev/rawctl", O_RDONLY);
	217	if (fd < 0) {
	218	fd = open("/dev/raw/rawctl", O_RDONLY);
	219	if (fd < 0)
8cc7afa9	220	return 1;
07e5b264 JA	221	}
	222
	223	rq.raw_minor = minor(dev);
	224	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	225	close(fd);
8cc7afa9	226	return 1;
07e5b264 JA	227	}
	228
	229	close(fd);
	230	*majdev = rq.block_major;
	231	*mindev = rq.block_minor;
8cc7afa9	232	return 0;
07e5b264 JA	233	}
07e5b264 JA	234
5b6f5c66	235	#ifdef O_NOATIME
5921e80c	236	#define FIO_O_NOATIME O_NOATIME
5b6f5c66 JA	237	#else
	238	#define FIO_O_NOATIME 0
	239	#endif
5921e80c	240
ebac4655	241	#endif