Merge branch 'master' into gfio
[fio.git] / os / os-linux.h
CommitLineData
ebac4655
JA
1#ifndef FIO_OS_LINUX_H
2#define FIO_OS_LINUX_H
3
cca84643
JA
4#define FIO_OS os_linux
5
ebac4655 6#include <sys/ioctl.h>
8756e4d4 7#include <sys/uio.h>
3c39a379 8#include <sys/syscall.h>
2e3bd4c2 9#include <sys/vfs.h>
3c39a379 10#include <unistd.h>
97fbdfa1 11#include <fcntl.h>
3c2d93ed 12#include <errno.h>
d09913c2 13#include <sched.h>
ea421790 14#include <linux/unistd.h>
07e5b264
JA
15#include <linux/raw.h>
16#include <linux/major.h>
610a730c 17#include <endian.h>
ebac4655 18
f3de88a7 19#include "indirect.h"
79a43187 20#include "binject.h"
e2e58886 21#include "../file.h"
f3de88a7 22
ebac4655
JA
23#define FIO_HAVE_LIBAIO
24#define FIO_HAVE_POSIXAIO
25#define FIO_HAVE_FADVISE
26#define FIO_HAVE_CPU_AFFINITY
27#define FIO_HAVE_DISK_UTIL
28#define FIO_HAVE_SGIO
ba4f8923 29#define FIO_HAVE_IOPRIO
8756e4d4 30#define FIO_HAVE_SPLICE
22f78b32 31#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28 32#define FIO_HAVE_ODIRECT
74b025b0 33#define FIO_HAVE_HUGETLB
07e5b264 34#define FIO_HAVE_RAWBIND
5e62c22a 35#define FIO_HAVE_BLKTRACE
5921e80c 36#define FIO_HAVE_STRSEP
207cb0f0 37#define FIO_HAVE_POSIXAIO_FSYNC
f356d01d 38#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38 39#define FIO_HAVE_CL_SIZE
a696fa2a 40#define FIO_HAVE_CGROUPS
c36d16f5 41#define FIO_HAVE_FDATASYNC
2e3bd4c2 42#define FIO_HAVE_FS_STAT
a5f3027c 43#define FIO_HAVE_TRIM
79a43187 44#define FIO_HAVE_BINJECT
ecc314ba 45#define FIO_HAVE_CLOCK_MONOTONIC
47f767c1 46#define FIO_HAVE_GETTID
93bcfd20 47#define FIO_USE_GENERIC_INIT_RANDOM_STATE
1ecc95ca 48#define FIO_HAVE_E4_ENG
b6959b55 49
804bb9a1
JA
50/*
51 * Can only enable this for newer glibcs, or the header and defines are
52 * missing
53 */
52ad4b94
JA
54#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
55#define FIO_HAVE_FALLOCATE
56#endif
78b269e3 57#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
804bb9a1
JA
58#define FIO_HAVE_LINUX_FALLOCATE
59#endif
60
1ecc95ca
JA
61#ifdef FIO_HAVE_LINUX_FALLOCATE
62#define FIO_HAVE_FALLOC_ENG
63#endif
64
093729f7
JA
65#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
66#define FIO_HAVE_SYNC_FILE_RANGE
67#endif
68
dc873b6f 69#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 70
4d8947de
SL
71#ifndef CLOCK_MONOTONIC
72#define CLOCK_MONOTONIC 1
73#endif
74
ebac4655 75typedef cpu_set_t os_cpu_mask_t;
4e78e405 76
6dfd46b9 77typedef struct drand48_data os_random_state_t;
ebac4655
JA
78
79/*
80 * we want fadvise64 really, but it's so tangled... later
81 */
4d8947de 82#ifdef FIO_HAVE_FADVISE
ebac4655
JA
83#define fadvise(fd, off, len, advice) \
84 posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de 85#endif
ebac4655 86
c8f025f8
JA
87/*
88 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
89 * the affinity helpers to work.
90 */
91#ifndef GLIBC_2_3_2
e8462bd8
JA
92#define fio_setaffinity(pid, cpumask) \
93 sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655
JA
94#define fio_getaffinity(pid, ptr) \
95 sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8 96#else
e8462bd8
JA
97#define fio_setaffinity(pid, cpumask) \
98 sched_setaffinity((pid), &(cpumask))
c8f025f8
JA
99#define fio_getaffinity(pid, ptr) \
100 sched_getaffinity((pid), (ptr))
101#endif
be4ecfdf 102
1f4c0a4f
JA
103#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
104#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
d2ce18b5
JA
105
106static inline int fio_cpuset_init(os_cpu_mask_t *mask)
107{
108 CPU_ZERO(mask);
109 return 0;
110}
111
112static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
113{
114 return 0;
115}
6d459ee7
JA
116
117#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de 118
28727df7
JA
119enum {
120 IOPRIO_CLASS_NONE,
121 IOPRIO_CLASS_RT,
122 IOPRIO_CLASS_BE,
123 IOPRIO_CLASS_IDLE,
124};
125
126enum {
127 IOPRIO_WHO_PROCESS = 1,
128 IOPRIO_WHO_PGRP,
129 IOPRIO_WHO_USER,
130};
131
132#define IOPRIO_BITS 16
133#define IOPRIO_CLASS_SHIFT 13
134
135static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
ebac4655 136{
28727df7
JA
137 /*
138 * If no class is set, assume BE
139 */
140 if (!ioprio_class)
141 ioprio_class = IOPRIO_CLASS_BE;
142
143 ioprio |= ioprio_class << IOPRIO_CLASS_SHIFT;
ebac4655
JA
144 return syscall(__NR_ioprio_set, which, who, ioprio);
145}
146
47f767c1
JA
147static inline int gettid(void)
148{
149 return syscall(__NR_gettid);
150}
151
97fbdfa1
JA
152/*
153 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
154 * aren't either.
155 */
156#ifndef SPLICE_F_MOVE
157#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
158#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
159 /* we may still block on the fd we splice */
160 /* from/to, of course */
161#define SPLICE_F_MORE (0x04) /* expect more data */
162#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4
JA
163
164static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
495ee9b6 165 size_t len, unsigned int flags)
8756e4d4 166{
97fbdfa1 167 return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4
JA
168}
169
170static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
171{
97fbdfa1 172 return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4
JA
173}
174
175static inline int vmsplice(int fd, const struct iovec *iov,
176 unsigned long nr_segs, unsigned int flags)
177{
97fbdfa1 178 return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4 179}
97fbdfa1 180#endif
8756e4d4 181
3feedc60
JA
182#define SPLICE_DEF_SIZE (64*1024)
183
b6959b55 184#ifdef FIO_HAVE_SYSLET
bf0dc8fa
IM
185
186struct syslet_uatom;
187struct async_head_user;
188
a4f4fdd7
JA
189/*
190 * syslet stuff
191 */
bf0dc8fa
IM
192static inline struct syslet_uatom *
193async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
a4f4fdd7 194{
2b52511f 195 return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7
JA
196}
197
bf0dc8fa
IM
198static inline long
199async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
200 struct async_head_user *ahu)
a4f4fdd7 201{
bf0dc8fa
IM
202 return syscall(__NR_async_wait, min_wait_events,
203 user_ring_idx, ahu);
a4f4fdd7
JA
204}
205
7756b0d0 206static inline long async_thread(void *event, struct async_head_user *ahu)
a4f4fdd7 207{
7756b0d0 208 return syscall(__NR_async_thread, event, ahu);
a4f4fdd7
JA
209}
210
211static inline long umem_add(unsigned long *uptr, unsigned long inc)
212{
213 return syscall(__NR_umem_add, uptr, inc);
214}
b6959b55 215#endif /* FIO_HAVE_SYSLET */
a4f4fdd7 216
ebac4655
JA
217#ifndef BLKGETSIZE64
218#define BLKGETSIZE64 _IOR(0x12,114,size_t)
219#endif
220
e5b401d4
JA
221#ifndef BLKFLSBUF
222#define BLKFLSBUF _IO(0x12,97)
223#endif
224
a5f3027c
JA
225#ifndef BLKDISCARD
226#define BLKDISCARD _IO(0x12,119)
227#endif
228
e2e58886 229static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4 230{
ecc314ba 231 return ioctl(f->fd, BLKFLSBUF);
e5b401d4
JA
232}
233
ecc314ba 234static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
ebac4655 235{
ecc314ba 236 if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655
JA
237 return 0;
238
239 return errno;
240}
241
32cd46a0
JA
242static inline unsigned long long os_phys_mem(void)
243{
244 long pagesize, pages;
245
246 pagesize = sysconf(_SC_PAGESIZE);
247 pages = sysconf(_SC_PHYS_PAGES);
248 if (pages == -1 || pagesize == -1)
249 return 0;
250
251 return (unsigned long long) pages * (unsigned long long) pagesize;
252}
253
6dfd46b9
JA
254static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
255{
256 srand48_r(seed, rs);
257}
258
259static inline long os_random_long(os_random_state_t *rs)
260{
261 long val;
262
263 lrand48_r(rs, &val);
264 return val;
265}
266
8cc7afa9 267static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
07e5b264
JA
268{
269 struct raw_config_request rq;
270 int fd;
271
272 if (major(dev) != RAW_MAJOR)
8cc7afa9 273 return 1;
07e5b264
JA
274
275 /*
276 * we should be able to find /dev/rawctl or /dev/raw/rawctl
277 */
278 fd = open("/dev/rawctl", O_RDONLY);
279 if (fd < 0) {
280 fd = open("/dev/raw/rawctl", O_RDONLY);
281 if (fd < 0)
8cc7afa9 282 return 1;
07e5b264
JA
283 }
284
285 rq.raw_minor = minor(dev);
286 if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
287 close(fd);
8cc7afa9 288 return 1;
07e5b264
JA
289 }
290
291 close(fd);
292 *majdev = rq.block_major;
293 *mindev = rq.block_minor;
8cc7afa9 294 return 0;
07e5b264
JA
295}
296
5b6f5c66 297#ifdef O_NOATIME
5921e80c 298#define FIO_O_NOATIME O_NOATIME
5b6f5c66
JA
299#else
300#define FIO_O_NOATIME 0
301#endif
5921e80c 302
a1c58075
JA
303#ifdef MADV_REMOVE
304#define FIO_MADV_FREE MADV_REMOVE
305#endif
306
610a730c
JA
307#if __BYTE_ORDER == __LITTLE_ENDIAN
308#define FIO_LITTLE_ENDIAN
309#elif __BYTE_ORDER == __BIG_ENDIAN
310#define FIO_BIG_ENDIAN
311#else
312#error "Unknown endianness"
313#endif
314
ff245192
JA
315#define fio_swap16(x) __bswap_16(x)
316#define fio_swap32(x) __bswap_32(x)
317#define fio_swap64(x) __bswap_64(x)
318
eb7ccf38
JA
319#define CACHE_LINE_FILE \
320 "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
321
322static inline int arch_cache_line_size(void)
323{
324 char size[32];
325 int fd, ret;
326
327 fd = open(CACHE_LINE_FILE, O_RDONLY);
328 if (fd < 0)
329 return -1;
330
331 ret = read(fd, size, sizeof(size));
ec76f15a
JA
332
333 close(fd);
334
eb7ccf38
JA
335 if (ret <= 0)
336 return -1;
ec76f15a
JA
337 else
338 return atoi(size);
eb7ccf38
JA
339}
340
2e3bd4c2
JA
341static inline unsigned long long get_fs_size(const char *path)
342{
343 unsigned long long ret;
344 struct statfs s;
345
346 if (statfs(path, &s) < 0)
347 return -1ULL;
348
349 ret = s.f_bsize;
350 ret *= (unsigned long long) s.f_bfree;
351 return ret;
352}
353
a5f3027c
JA
354static inline int os_trim(int fd, unsigned long long start,
355 unsigned long long len)
356{
357 uint64_t range[2];
358
359 range[0] = start;
360 range[1] = len;
361
362 if (!ioctl(fd, BLKDISCARD, range))
363 return 0;
364
365 return errno;
366}
367
ebac4655 368#endif