[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 <sys/vfs.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 "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_FALLOCATE
#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

/*
 * Can only enable this for newer glibcs, or the header and defines are
 * missing
 */
#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
#define FIO_HAVE_LINUX_FALLOCATE
#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)	CPU_CLR((cpu), (mask))
#define fio_cpu_set(mask, cpu)		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

#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
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	6	#include <sys/syscall.h>
2e3bd4c2	7	#include <sys/vfs.h>
3c39a379	8	#include <unistd.h>
97fbdfa1	9	#include <fcntl.h>
3c2d93ed	10	#include <errno.h>
d09913c2	11	#include <sched.h>
ea421790	12	#include <linux/unistd.h>
07e5b264 JA	13	#include <linux/raw.h>
07e5b264 JA	14	#include <linux/major.h>
ebac4655	15
f3de88a7	16	#include "indirect.h"
79a43187	17	#include "binject.h"
e2e58886	18	#include "../file.h"
f3de88a7	19
ebac4655 JA	20	#define FIO_HAVE_LIBAIO
	21	#define FIO_HAVE_POSIXAIO
	22	#define FIO_HAVE_FADVISE
	23	#define FIO_HAVE_CPU_AFFINITY
	24	#define FIO_HAVE_DISK_UTIL
	25	#define FIO_HAVE_SGIO
ba4f8923	26	#define FIO_HAVE_IOPRIO
8756e4d4	27	#define FIO_HAVE_SPLICE
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
5921e80c	33	#define FIO_HAVE_STRSEP
fffca02d	34	#define FIO_HAVE_FALLOCATE
207cb0f0	35	#define FIO_HAVE_POSIXAIO_FSYNC
f356d01d	36	#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38	37	#define FIO_HAVE_CL_SIZE
a696fa2a	38	#define FIO_HAVE_CGROUPS
c36d16f5	39	#define FIO_HAVE_FDATASYNC
2e3bd4c2	40	#define FIO_HAVE_FS_STAT
a5f3027c	41	#define FIO_HAVE_TRIM
79a43187	42	#define FIO_HAVE_BINJECT
ecc314ba	43	#define FIO_HAVE_CLOCK_MONOTONIC
47f767c1	44	#define FIO_HAVE_GETTID
b6959b55	45
804bb9a1 JA	46	/*
	47	* Can only enable this for newer glibcs, or the header and defines are
	48	* missing
	49	*/
	50	#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
	51	#define FIO_HAVE_LINUX_FALLOCATE
	52	#endif
	53
093729f7 JA	54	#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
	55	#define FIO_HAVE_SYNC_FILE_RANGE
	56	#endif
	57
dc873b6f	58	#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655	59
4d8947de SL	60	#ifndef CLOCK_MONOTONIC
	61	#define CLOCK_MONOTONIC 1
	62	#endif
	63
ebac4655	64	typedef cpu_set_t os_cpu_mask_t;
4e78e405	65
6dfd46b9	66	typedef struct drand48_data os_random_state_t;
ebac4655 JA	67
	68	/*
	69	* we want fadvise64 really, but it's so tangled... later
	70	*/
4d8947de	71	#ifdef FIO_HAVE_FADVISE
ebac4655 JA	72	#define fadvise(fd, off, len, advice) \
ebac4655 JA	73	posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de	74	#endif
ebac4655	75
c8f025f8 JA	76	/*
	77	* If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
	78	* the affinity helpers to work.
	79	*/
	80	#ifndef GLIBC_2_3_2
e8462bd8 JA	81	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	82	sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655 JA	83	#define fio_getaffinity(pid, ptr) \
ebac4655 JA	84	sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8	85	#else
e8462bd8 JA	86	#define fio_setaffinity(pid, cpumask) \
e8462bd8 JA	87	sched_setaffinity((pid), &(cpumask))
c8f025f8 JA	88	#define fio_getaffinity(pid, ptr) \
	89	sched_getaffinity((pid), (ptr))
	90	#endif
be4ecfdf JA	91
be4ecfdf JA	92	#define fio_cpu_clear(mask, cpu) CPU_CLR((cpu), (mask))
6d459ee7	93	#define fio_cpu_set(mask, cpu) CPU_SET((cpu), (mask))
d2ce18b5 JA	94
	95	static inline int fio_cpuset_init(os_cpu_mask_t *mask)
	96	{
	97	CPU_ZERO(mask);
	98	return 0;
	99	}
	100
	101	static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
	102	{
	103	return 0;
	104	}
6d459ee7 JA	105
6d459ee7 JA	106	#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de	107
ebac4655 JA	108	static inline int ioprio_set(int which, int who, int ioprio)
	109	{
	110	return syscall(__NR_ioprio_set, which, who, ioprio);
	111	}
	112
47f767c1 JA	113	static inline int gettid(void)
	114	{
	115	return syscall(__NR_gettid);
	116	}
	117
97fbdfa1 JA	118	/*
	119	* Just check for SPLICE_F_MOVE, if that isn't there, assume the others
	120	* aren't either.
	121	*/
	122	#ifndef SPLICE_F_MOVE
	123	#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
	124	#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
	125	/* we may still block on the fd we splice */
	126	/* from/to, of course */
	127	#define SPLICE_F_MORE (0x04) /* expect more data */
	128	#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4 JA	129
8756e4d4 JA	130	static inline int splice(int fdin, loff_t off_in, int fdout, loff_t off_out,
495ee9b6	131	size_t len, unsigned int flags)
8756e4d4	132	{
97fbdfa1	133	return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4 JA	134	}
	135
	136	static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
	137	{
97fbdfa1	138	return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4 JA	139	}
	140
	141	static inline int vmsplice(int fd, const struct iovec *iov,
	142	unsigned long nr_segs, unsigned int flags)
	143	{
97fbdfa1	144	return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4	145	}
97fbdfa1	146	#endif
8756e4d4	147
3feedc60 JA	148	#define SPLICE_DEF_SIZE (64*1024)
3feedc60 JA	149
b6959b55	150	#ifdef FIO_HAVE_SYSLET
bf0dc8fa IM	151
	152	struct syslet_uatom;
	153	struct async_head_user;
	154
a4f4fdd7 JA	155	/*
	156	* syslet stuff
	157	*/
bf0dc8fa IM	158	static inline struct syslet_uatom *
bf0dc8fa IM	159	async_exec(struct syslet_uatom atom, struct async_head_user ahu)
a4f4fdd7	160	{
2b52511f	161	return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7 JA	162	}
a4f4fdd7 JA	163
bf0dc8fa IM	164	static inline long
	165	async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
	166	struct async_head_user *ahu)
a4f4fdd7	167	{
bf0dc8fa IM	168	return syscall(__NR_async_wait, min_wait_events,
bf0dc8fa IM	169	user_ring_idx, ahu);
a4f4fdd7 JA	170	}
a4f4fdd7 JA	171
7756b0d0	172	static inline long async_thread(void event, struct async_head_user ahu)
a4f4fdd7	173	{
7756b0d0	174	return syscall(__NR_async_thread, event, ahu);
a4f4fdd7 JA	175	}
	176
	177	static inline long umem_add(unsigned long *uptr, unsigned long inc)
	178	{
	179	return syscall(__NR_umem_add, uptr, inc);
	180	}
b6959b55	181	#endif /* FIO_HAVE_SYSLET */
a4f4fdd7	182
6cefbe33 JA	183	enum {
	184	IOPRIO_CLASS_NONE,
	185	IOPRIO_CLASS_RT,
	186	IOPRIO_CLASS_BE,
	187	IOPRIO_CLASS_IDLE,
	188	};
	189
ebac4655 JA	190	enum {
	191	IOPRIO_WHO_PROCESS = 1,
	192	IOPRIO_WHO_PGRP,
	193	IOPRIO_WHO_USER,
	194	};
	195
6cefbe33	196	#define IOPRIO_BITS 16
ebac4655 JA	197	#define IOPRIO_CLASS_SHIFT 13
	198
	199	#ifndef BLKGETSIZE64
	200	#define BLKGETSIZE64 _IOR(0x12,114,size_t)
	201	#endif
	202
e5b401d4 JA	203	#ifndef BLKFLSBUF
	204	#define BLKFLSBUF _IO(0x12,97)
	205	#endif
	206
a5f3027c JA	207	#ifndef BLKDISCARD
	208	#define BLKDISCARD _IO(0x12,119)
	209	#endif
	210
e2e58886	211	static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4	212	{
ecc314ba	213	return ioctl(f->fd, BLKFLSBUF);
e5b401d4 JA	214	}
e5b401d4 JA	215
ecc314ba	216	static inline int blockdev_size(struct fio_file f, unsigned long long bytes)
ebac4655	217	{
ecc314ba	218	if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655 JA	219	return 0;
	220
	221	return errno;
	222	}
	223
32cd46a0 JA	224	static inline unsigned long long os_phys_mem(void)
	225	{
	226	long pagesize, pages;
	227
	228	pagesize = sysconf(_SC_PAGESIZE);
	229	pages = sysconf(_SC_PHYS_PAGES);
	230	if (pages == -1 \|\| pagesize == -1)
	231	return 0;
	232
	233	return (unsigned long long) pages * (unsigned long long) pagesize;
	234	}
	235
6dfd46b9 JA	236	static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
	237	{
	238	srand48_r(seed, rs);
	239	}
	240
	241	static inline long os_random_long(os_random_state_t *rs)
	242	{
	243	long val;
	244
	245	lrand48_r(rs, &val);
	246	return val;
	247	}
	248
8cc7afa9	249	static inline int fio_lookup_raw(dev_t dev, int majdev, int mindev)
07e5b264 JA	250	{
	251	struct raw_config_request rq;
	252	int fd;
	253
	254	if (major(dev) != RAW_MAJOR)
8cc7afa9	255	return 1;
07e5b264 JA	256
	257	/*
	258	* we should be able to find /dev/rawctl or /dev/raw/rawctl
	259	*/
	260	fd = open("/dev/rawctl", O_RDONLY);
	261	if (fd < 0) {
	262	fd = open("/dev/raw/rawctl", O_RDONLY);
	263	if (fd < 0)
8cc7afa9	264	return 1;
07e5b264 JA	265	}
	266
	267	rq.raw_minor = minor(dev);
	268	if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
	269	close(fd);
8cc7afa9	270	return 1;
07e5b264 JA	271	}
	272
	273	close(fd);
	274	*majdev = rq.block_major;
	275	*mindev = rq.block_minor;
8cc7afa9	276	return 0;
07e5b264 JA	277	}
07e5b264 JA	278
5b6f5c66	279	#ifdef O_NOATIME
5921e80c	280	#define FIO_O_NOATIME O_NOATIME
5b6f5c66 JA	281	#else
	282	#define FIO_O_NOATIME 0
	283	#endif
5921e80c	284
a1c58075 JA	285	#ifdef MADV_REMOVE
	286	#define FIO_MADV_FREE MADV_REMOVE
	287	#endif
	288
eb7ccf38 JA	289	#define CACHE_LINE_FILE \
	290	"/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
	291
	292	static inline int arch_cache_line_size(void)
	293	{
	294	char size[32];
	295	int fd, ret;
	296
	297	fd = open(CACHE_LINE_FILE, O_RDONLY);
	298	if (fd < 0)
	299	return -1;
	300
	301	ret = read(fd, size, sizeof(size));
ec76f15a JA	302
	303	close(fd);
	304
eb7ccf38 JA	305	if (ret <= 0)
eb7ccf38 JA	306	return -1;
ec76f15a JA	307	else
ec76f15a JA	308	return atoi(size);
eb7ccf38 JA	309	}
eb7ccf38 JA	310
2e3bd4c2 JA	311	static inline unsigned long long get_fs_size(const char *path)
	312	{
	313	unsigned long long ret;
	314	struct statfs s;
	315
	316	if (statfs(path, &s) < 0)
	317	return -1ULL;
	318
	319	ret = s.f_bsize;
	320	ret *= (unsigned long long) s.f_bfree;
	321	return ret;
	322	}
	323
a5f3027c JA	324	static inline int os_trim(int fd, unsigned long long start,
	325	unsigned long long len)
	326	{
	327	uint64_t range[2];
	328
	329	range[0] = start;
	330	range[1] = len;
	331
	332	if (!ioctl(fd, BLKDISCARD, range))
	333	return 0;
	334
	335	return errno;
	336	}
	337
ebac4655	338	#endif