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>
39b93568 10#include <sys/mman.h>
3c39a379 11#include <unistd.h>
97fbdfa1 12#include <fcntl.h>
3c2d93ed 13#include <errno.h>
d09913c2 14#include <sched.h>
ea421790 15#include <linux/unistd.h>
07e5b264
JA
16#include <linux/raw.h>
17#include <linux/major.h>
610a730c 18#include <endian.h>
ebac4655 19
f3de88a7 20#include "indirect.h"
79a43187 21#include "binject.h"
e2e58886 22#include "../file.h"
f3de88a7 23
ebac4655
JA
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
ba4f8923 30#define FIO_HAVE_IOPRIO
8756e4d4 31#define FIO_HAVE_SPLICE
22f78b32 32#define FIO_HAVE_IOSCHED_SWITCH
2c0ecd28 33#define FIO_HAVE_ODIRECT
74b025b0 34#define FIO_HAVE_HUGETLB
07e5b264 35#define FIO_HAVE_RAWBIND
5e62c22a 36#define FIO_HAVE_BLKTRACE
5921e80c 37#define FIO_HAVE_STRSEP
207cb0f0 38#define FIO_HAVE_POSIXAIO_FSYNC
f356d01d 39#define FIO_HAVE_PSHARED_MUTEX
eb7ccf38 40#define FIO_HAVE_CL_SIZE
a696fa2a 41#define FIO_HAVE_CGROUPS
c36d16f5 42#define FIO_HAVE_FDATASYNC
2e3bd4c2 43#define FIO_HAVE_FS_STAT
a5f3027c 44#define FIO_HAVE_TRIM
79a43187 45#define FIO_HAVE_BINJECT
ecc314ba 46#define FIO_HAVE_CLOCK_MONOTONIC
47f767c1 47#define FIO_HAVE_GETTID
93bcfd20 48#define FIO_USE_GENERIC_INIT_RANDOM_STATE
1ecc95ca 49#define FIO_HAVE_E4_ENG
b6959b55 50
39b93568
JA
51#ifdef MAP_HUGETLB
52#define FIO_HAVE_MMAP_HUGE
53#endif
54
804bb9a1
JA
55/*
56 * Can only enable this for newer glibcs, or the header and defines are
57 * missing
58 */
52ad4b94
JA
59#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
60#define FIO_HAVE_FALLOCATE
61#endif
78b269e3 62#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
804bb9a1
JA
63#define FIO_HAVE_LINUX_FALLOCATE
64#endif
65
1ecc95ca
JA
66#ifdef FIO_HAVE_LINUX_FALLOCATE
67#define FIO_HAVE_FALLOC_ENG
68#endif
69
093729f7
JA
70#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
71#define FIO_HAVE_SYNC_FILE_RANGE
72#endif
73
dc873b6f 74#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 75
4d8947de
SL
76#ifndef CLOCK_MONOTONIC
77#define CLOCK_MONOTONIC 1
78#endif
79
ebac4655 80typedef cpu_set_t os_cpu_mask_t;
4e78e405 81
6dfd46b9 82typedef struct drand48_data os_random_state_t;
ebac4655
JA
83
84/*
85 * we want fadvise64 really, but it's so tangled... later
86 */
4d8947de 87#ifdef FIO_HAVE_FADVISE
ebac4655
JA
88#define fadvise(fd, off, len, advice) \
89 posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de 90#endif
ebac4655 91
c8f025f8
JA
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
e8462bd8
JA
97#define fio_setaffinity(pid, cpumask) \
98 sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655
JA
99#define fio_getaffinity(pid, ptr) \
100 sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8 101#else
e8462bd8
JA
102#define fio_setaffinity(pid, cpumask) \
103 sched_setaffinity((pid), &(cpumask))
c8f025f8
JA
104#define fio_getaffinity(pid, ptr) \
105 sched_getaffinity((pid), (ptr))
106#endif
be4ecfdf 107
1f4c0a4f
JA
108#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
109#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
d2ce18b5
JA
110
111static inline int fio_cpuset_init(os_cpu_mask_t *mask)
112{
113 CPU_ZERO(mask);
114 return 0;
115}
116
117static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
118{
119 return 0;
120}
6d459ee7
JA
121
122#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de 123
28727df7
JA
124enum {
125 IOPRIO_CLASS_NONE,
126 IOPRIO_CLASS_RT,
127 IOPRIO_CLASS_BE,
128 IOPRIO_CLASS_IDLE,
129};
130
131enum {
132 IOPRIO_WHO_PROCESS = 1,
133 IOPRIO_WHO_PGRP,
134 IOPRIO_WHO_USER,
135};
136
137#define IOPRIO_BITS 16
138#define IOPRIO_CLASS_SHIFT 13
139
140static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
ebac4655 141{
28727df7
JA
142 /*
143 * If no class is set, assume BE
144 */
145 if (!ioprio_class)
146 ioprio_class = IOPRIO_CLASS_BE;
147
148 ioprio |= ioprio_class << IOPRIO_CLASS_SHIFT;
ebac4655
JA
149 return syscall(__NR_ioprio_set, which, who, ioprio);
150}
151
47f767c1
JA
152static inline int gettid(void)
153{
154 return syscall(__NR_gettid);
155}
156
97fbdfa1
JA
157/*
158 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
159 * aren't either.
160 */
161#ifndef SPLICE_F_MOVE
162#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
163#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
164 /* we may still block on the fd we splice */
165 /* from/to, of course */
166#define SPLICE_F_MORE (0x04) /* expect more data */
167#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4
JA
168
169static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
495ee9b6 170 size_t len, unsigned int flags)
8756e4d4 171{
97fbdfa1 172 return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4
JA
173}
174
175static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
176{
97fbdfa1 177 return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4
JA
178}
179
180static inline int vmsplice(int fd, const struct iovec *iov,
181 unsigned long nr_segs, unsigned int flags)
182{
97fbdfa1 183 return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4 184}
97fbdfa1 185#endif
8756e4d4 186
3feedc60
JA
187#define SPLICE_DEF_SIZE (64*1024)
188
b6959b55 189#ifdef FIO_HAVE_SYSLET
bf0dc8fa
IM
190
191struct syslet_uatom;
192struct async_head_user;
193
a4f4fdd7
JA
194/*
195 * syslet stuff
196 */
bf0dc8fa
IM
197static inline struct syslet_uatom *
198async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
a4f4fdd7 199{
2b52511f 200 return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7
JA
201}
202
bf0dc8fa
IM
203static inline long
204async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
205 struct async_head_user *ahu)
a4f4fdd7 206{
bf0dc8fa
IM
207 return syscall(__NR_async_wait, min_wait_events,
208 user_ring_idx, ahu);
a4f4fdd7
JA
209}
210
7756b0d0 211static inline long async_thread(void *event, struct async_head_user *ahu)
a4f4fdd7 212{
7756b0d0 213 return syscall(__NR_async_thread, event, ahu);
a4f4fdd7
JA
214}
215
216static inline long umem_add(unsigned long *uptr, unsigned long inc)
217{
218 return syscall(__NR_umem_add, uptr, inc);
219}
b6959b55 220#endif /* FIO_HAVE_SYSLET */
a4f4fdd7 221
ebac4655
JA
222#ifndef BLKGETSIZE64
223#define BLKGETSIZE64 _IOR(0x12,114,size_t)
224#endif
225
e5b401d4
JA
226#ifndef BLKFLSBUF
227#define BLKFLSBUF _IO(0x12,97)
228#endif
229
a5f3027c
JA
230#ifndef BLKDISCARD
231#define BLKDISCARD _IO(0x12,119)
232#endif
233
e2e58886 234static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4 235{
ecc314ba 236 return ioctl(f->fd, BLKFLSBUF);
e5b401d4
JA
237}
238
ecc314ba 239static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
ebac4655 240{
ecc314ba 241 if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655
JA
242 return 0;
243
244 return errno;
245}
246
32cd46a0
JA
247static inline unsigned long long os_phys_mem(void)
248{
249 long pagesize, pages;
250
251 pagesize = sysconf(_SC_PAGESIZE);
252 pages = sysconf(_SC_PHYS_PAGES);
253 if (pages == -1 || pagesize == -1)
254 return 0;
255
256 return (unsigned long long) pages * (unsigned long long) pagesize;
257}
258
6dfd46b9
JA
259static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
260{
261 srand48_r(seed, rs);
262}
263
264static inline long os_random_long(os_random_state_t *rs)
265{
266 long val;
267
268 lrand48_r(rs, &val);
269 return val;
270}
271
8cc7afa9 272static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
07e5b264
JA
273{
274 struct raw_config_request rq;
275 int fd;
276
277 if (major(dev) != RAW_MAJOR)
8cc7afa9 278 return 1;
07e5b264
JA
279
280 /*
281 * we should be able to find /dev/rawctl or /dev/raw/rawctl
282 */
283 fd = open("/dev/rawctl", O_RDONLY);
284 if (fd < 0) {
285 fd = open("/dev/raw/rawctl", O_RDONLY);
286 if (fd < 0)
8cc7afa9 287 return 1;
07e5b264
JA
288 }
289
290 rq.raw_minor = minor(dev);
291 if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
292 close(fd);
8cc7afa9 293 return 1;
07e5b264
JA
294 }
295
296 close(fd);
297 *majdev = rq.block_major;
298 *mindev = rq.block_minor;
8cc7afa9 299 return 0;
07e5b264
JA
300}
301
5b6f5c66 302#ifdef O_NOATIME
5921e80c 303#define FIO_O_NOATIME O_NOATIME
5b6f5c66
JA
304#else
305#define FIO_O_NOATIME 0
306#endif
5921e80c 307
a1c58075
JA
308#ifdef MADV_REMOVE
309#define FIO_MADV_FREE MADV_REMOVE
310#endif
311
610a730c
JA
312#if __BYTE_ORDER == __LITTLE_ENDIAN
313#define FIO_LITTLE_ENDIAN
314#elif __BYTE_ORDER == __BIG_ENDIAN
315#define FIO_BIG_ENDIAN
316#else
317#error "Unknown endianness"
318#endif
319
ff245192
JA
320#define fio_swap16(x) __bswap_16(x)
321#define fio_swap32(x) __bswap_32(x)
322#define fio_swap64(x) __bswap_64(x)
323
eb7ccf38
JA
324#define CACHE_LINE_FILE \
325 "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
326
327static inline int arch_cache_line_size(void)
328{
329 char size[32];
330 int fd, ret;
331
332 fd = open(CACHE_LINE_FILE, O_RDONLY);
333 if (fd < 0)
334 return -1;
335
336 ret = read(fd, size, sizeof(size));
ec76f15a
JA
337
338 close(fd);
339
eb7ccf38
JA
340 if (ret <= 0)
341 return -1;
ec76f15a
JA
342 else
343 return atoi(size);
eb7ccf38
JA
344}
345
2e3bd4c2
JA
346static inline unsigned long long get_fs_size(const char *path)
347{
348 unsigned long long ret;
349 struct statfs s;
350
351 if (statfs(path, &s) < 0)
352 return -1ULL;
353
354 ret = s.f_bsize;
355 ret *= (unsigned long long) s.f_bfree;
356 return ret;
357}
358
a5f3027c
JA
359static inline int os_trim(int fd, unsigned long long start,
360 unsigned long long len)
361{
362 uint64_t range[2];
363
364 range[0] = start;
365 range[1] = len;
366
367 if (!ioctl(fd, BLKDISCARD, range))
368 return 0;
369
370 return errno;
371}
372
ebac4655 373#endif