Commit | Line | Data |
---|---|---|
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 |
b6959b55 | 47 | |
804bb9a1 JA |
48 | /* |
49 | * Can only enable this for newer glibcs, or the header and defines are | |
50 | * missing | |
51 | */ | |
78b269e3 | 52 | #if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8 |
804bb9a1 JA |
53 | #define FIO_HAVE_LINUX_FALLOCATE |
54 | #endif | |
55 | ||
093729f7 JA |
56 | #ifdef SYNC_FILE_RANGE_WAIT_BEFORE |
57 | #define FIO_HAVE_SYNC_FILE_RANGE | |
58 | #endif | |
59 | ||
dc873b6f | 60 | #define OS_MAP_ANON MAP_ANONYMOUS |
ebac4655 | 61 | |
4d8947de SL |
62 | #ifndef CLOCK_MONOTONIC |
63 | #define CLOCK_MONOTONIC 1 | |
64 | #endif | |
65 | ||
ebac4655 | 66 | typedef cpu_set_t os_cpu_mask_t; |
4e78e405 | 67 | |
6dfd46b9 | 68 | typedef struct drand48_data os_random_state_t; |
ebac4655 JA |
69 | |
70 | /* | |
71 | * we want fadvise64 really, but it's so tangled... later | |
72 | */ | |
4d8947de | 73 | #ifdef FIO_HAVE_FADVISE |
ebac4655 JA |
74 | #define fadvise(fd, off, len, advice) \ |
75 | posix_fadvise((fd), (off_t)(off), (len), (advice)) | |
4d8947de | 76 | #endif |
ebac4655 | 77 | |
c8f025f8 JA |
78 | /* |
79 | * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want | |
80 | * the affinity helpers to work. | |
81 | */ | |
82 | #ifndef GLIBC_2_3_2 | |
e8462bd8 JA |
83 | #define fio_setaffinity(pid, cpumask) \ |
84 | sched_setaffinity((pid), sizeof(cpumask), &(cpumask)) | |
ebac4655 JA |
85 | #define fio_getaffinity(pid, ptr) \ |
86 | sched_getaffinity((pid), sizeof(cpu_set_t), (ptr)) | |
c8f025f8 | 87 | #else |
e8462bd8 JA |
88 | #define fio_setaffinity(pid, cpumask) \ |
89 | sched_setaffinity((pid), &(cpumask)) | |
c8f025f8 JA |
90 | #define fio_getaffinity(pid, ptr) \ |
91 | sched_getaffinity((pid), (ptr)) | |
92 | #endif | |
be4ecfdf | 93 | |
1f4c0a4f JA |
94 | #define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask)) |
95 | #define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask)) | |
d2ce18b5 JA |
96 | |
97 | static inline int fio_cpuset_init(os_cpu_mask_t *mask) | |
98 | { | |
99 | CPU_ZERO(mask); | |
100 | return 0; | |
101 | } | |
102 | ||
103 | static inline int fio_cpuset_exit(os_cpu_mask_t *mask) | |
104 | { | |
105 | return 0; | |
106 | } | |
6d459ee7 JA |
107 | |
108 | #define FIO_MAX_CPUS CPU_SETSIZE | |
4d8947de | 109 | |
ebac4655 JA |
110 | static inline int ioprio_set(int which, int who, int ioprio) |
111 | { | |
112 | return syscall(__NR_ioprio_set, which, who, ioprio); | |
113 | } | |
114 | ||
47f767c1 JA |
115 | static inline int gettid(void) |
116 | { | |
117 | return syscall(__NR_gettid); | |
118 | } | |
119 | ||
97fbdfa1 JA |
120 | /* |
121 | * Just check for SPLICE_F_MOVE, if that isn't there, assume the others | |
122 | * aren't either. | |
123 | */ | |
124 | #ifndef SPLICE_F_MOVE | |
125 | #define SPLICE_F_MOVE (0x01) /* move pages instead of copying */ | |
126 | #define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */ | |
127 | /* we may still block on the fd we splice */ | |
128 | /* from/to, of course */ | |
129 | #define SPLICE_F_MORE (0x04) /* expect more data */ | |
130 | #define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */ | |
8756e4d4 JA |
131 | |
132 | static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out, | |
495ee9b6 | 133 | size_t len, unsigned int flags) |
8756e4d4 | 134 | { |
97fbdfa1 | 135 | return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags); |
8756e4d4 JA |
136 | } |
137 | ||
138 | static inline int tee(int fdin, int fdout, size_t len, unsigned int flags) | |
139 | { | |
97fbdfa1 | 140 | return syscall(__NR_sys_tee, fdin, fdout, len, flags); |
8756e4d4 JA |
141 | } |
142 | ||
143 | static inline int vmsplice(int fd, const struct iovec *iov, | |
144 | unsigned long nr_segs, unsigned int flags) | |
145 | { | |
97fbdfa1 | 146 | return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags); |
8756e4d4 | 147 | } |
97fbdfa1 | 148 | #endif |
8756e4d4 | 149 | |
3feedc60 JA |
150 | #define SPLICE_DEF_SIZE (64*1024) |
151 | ||
b6959b55 | 152 | #ifdef FIO_HAVE_SYSLET |
bf0dc8fa IM |
153 | |
154 | struct syslet_uatom; | |
155 | struct async_head_user; | |
156 | ||
a4f4fdd7 JA |
157 | /* |
158 | * syslet stuff | |
159 | */ | |
bf0dc8fa IM |
160 | static inline struct syslet_uatom * |
161 | async_exec(struct syslet_uatom *atom, struct async_head_user *ahu) | |
a4f4fdd7 | 162 | { |
2b52511f | 163 | return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu); |
a4f4fdd7 JA |
164 | } |
165 | ||
bf0dc8fa IM |
166 | static inline long |
167 | async_wait(unsigned long min_wait_events, unsigned long user_ring_idx, | |
168 | struct async_head_user *ahu) | |
a4f4fdd7 | 169 | { |
bf0dc8fa IM |
170 | return syscall(__NR_async_wait, min_wait_events, |
171 | user_ring_idx, ahu); | |
a4f4fdd7 JA |
172 | } |
173 | ||
7756b0d0 | 174 | static inline long async_thread(void *event, struct async_head_user *ahu) |
a4f4fdd7 | 175 | { |
7756b0d0 | 176 | return syscall(__NR_async_thread, event, ahu); |
a4f4fdd7 JA |
177 | } |
178 | ||
179 | static inline long umem_add(unsigned long *uptr, unsigned long inc) | |
180 | { | |
181 | return syscall(__NR_umem_add, uptr, inc); | |
182 | } | |
b6959b55 | 183 | #endif /* FIO_HAVE_SYSLET */ |
a4f4fdd7 | 184 | |
6cefbe33 JA |
185 | enum { |
186 | IOPRIO_CLASS_NONE, | |
187 | IOPRIO_CLASS_RT, | |
188 | IOPRIO_CLASS_BE, | |
189 | IOPRIO_CLASS_IDLE, | |
190 | }; | |
191 | ||
ebac4655 JA |
192 | enum { |
193 | IOPRIO_WHO_PROCESS = 1, | |
194 | IOPRIO_WHO_PGRP, | |
195 | IOPRIO_WHO_USER, | |
196 | }; | |
197 | ||
6cefbe33 | 198 | #define IOPRIO_BITS 16 |
ebac4655 JA |
199 | #define IOPRIO_CLASS_SHIFT 13 |
200 | ||
201 | #ifndef BLKGETSIZE64 | |
202 | #define BLKGETSIZE64 _IOR(0x12,114,size_t) | |
203 | #endif | |
204 | ||
e5b401d4 JA |
205 | #ifndef BLKFLSBUF |
206 | #define BLKFLSBUF _IO(0x12,97) | |
207 | #endif | |
208 | ||
a5f3027c JA |
209 | #ifndef BLKDISCARD |
210 | #define BLKDISCARD _IO(0x12,119) | |
211 | #endif | |
212 | ||
e2e58886 | 213 | static inline int blockdev_invalidate_cache(struct fio_file *f) |
e5b401d4 | 214 | { |
ecc314ba | 215 | return ioctl(f->fd, BLKFLSBUF); |
e5b401d4 JA |
216 | } |
217 | ||
ecc314ba | 218 | static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes) |
ebac4655 | 219 | { |
ecc314ba | 220 | if (!ioctl(f->fd, BLKGETSIZE64, bytes)) |
ebac4655 JA |
221 | return 0; |
222 | ||
223 | return errno; | |
224 | } | |
225 | ||
32cd46a0 JA |
226 | static inline unsigned long long os_phys_mem(void) |
227 | { | |
228 | long pagesize, pages; | |
229 | ||
230 | pagesize = sysconf(_SC_PAGESIZE); | |
231 | pages = sysconf(_SC_PHYS_PAGES); | |
232 | if (pages == -1 || pagesize == -1) | |
233 | return 0; | |
234 | ||
235 | return (unsigned long long) pages * (unsigned long long) pagesize; | |
236 | } | |
237 | ||
6dfd46b9 JA |
238 | static inline void os_random_seed(unsigned long seed, os_random_state_t *rs) |
239 | { | |
240 | srand48_r(seed, rs); | |
241 | } | |
242 | ||
243 | static inline long os_random_long(os_random_state_t *rs) | |
244 | { | |
245 | long val; | |
246 | ||
247 | lrand48_r(rs, &val); | |
248 | return val; | |
249 | } | |
250 | ||
8cc7afa9 | 251 | static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev) |
07e5b264 JA |
252 | { |
253 | struct raw_config_request rq; | |
254 | int fd; | |
255 | ||
256 | if (major(dev) != RAW_MAJOR) | |
8cc7afa9 | 257 | return 1; |
07e5b264 JA |
258 | |
259 | /* | |
260 | * we should be able to find /dev/rawctl or /dev/raw/rawctl | |
261 | */ | |
262 | fd = open("/dev/rawctl", O_RDONLY); | |
263 | if (fd < 0) { | |
264 | fd = open("/dev/raw/rawctl", O_RDONLY); | |
265 | if (fd < 0) | |
8cc7afa9 | 266 | return 1; |
07e5b264 JA |
267 | } |
268 | ||
269 | rq.raw_minor = minor(dev); | |
270 | if (ioctl(fd, RAW_GETBIND, &rq) < 0) { | |
271 | close(fd); | |
8cc7afa9 | 272 | return 1; |
07e5b264 JA |
273 | } |
274 | ||
275 | close(fd); | |
276 | *majdev = rq.block_major; | |
277 | *mindev = rq.block_minor; | |
8cc7afa9 | 278 | return 0; |
07e5b264 JA |
279 | } |
280 | ||
5b6f5c66 | 281 | #ifdef O_NOATIME |
5921e80c | 282 | #define FIO_O_NOATIME O_NOATIME |
5b6f5c66 JA |
283 | #else |
284 | #define FIO_O_NOATIME 0 | |
285 | #endif | |
5921e80c | 286 | |
a1c58075 JA |
287 | #ifdef MADV_REMOVE |
288 | #define FIO_MADV_FREE MADV_REMOVE | |
289 | #endif | |
290 | ||
610a730c JA |
291 | #if __BYTE_ORDER == __LITTLE_ENDIAN |
292 | #define FIO_LITTLE_ENDIAN | |
293 | #elif __BYTE_ORDER == __BIG_ENDIAN | |
294 | #define FIO_BIG_ENDIAN | |
295 | #else | |
296 | #error "Unknown endianness" | |
297 | #endif | |
298 | ||
ff245192 JA |
299 | #define fio_swap16(x) __bswap_16(x) |
300 | #define fio_swap32(x) __bswap_32(x) | |
301 | #define fio_swap64(x) __bswap_64(x) | |
302 | ||
eb7ccf38 JA |
303 | #define CACHE_LINE_FILE \ |
304 | "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size" | |
305 | ||
306 | static inline int arch_cache_line_size(void) | |
307 | { | |
308 | char size[32]; | |
309 | int fd, ret; | |
310 | ||
311 | fd = open(CACHE_LINE_FILE, O_RDONLY); | |
312 | if (fd < 0) | |
313 | return -1; | |
314 | ||
315 | ret = read(fd, size, sizeof(size)); | |
ec76f15a JA |
316 | |
317 | close(fd); | |
318 | ||
eb7ccf38 JA |
319 | if (ret <= 0) |
320 | return -1; | |
ec76f15a JA |
321 | else |
322 | return atoi(size); | |
eb7ccf38 JA |
323 | } |
324 | ||
2e3bd4c2 JA |
325 | static inline unsigned long long get_fs_size(const char *path) |
326 | { | |
327 | unsigned long long ret; | |
328 | struct statfs s; | |
329 | ||
330 | if (statfs(path, &s) < 0) | |
331 | return -1ULL; | |
332 | ||
333 | ret = s.f_bsize; | |
334 | ret *= (unsigned long long) s.f_bfree; | |
335 | return ret; | |
336 | } | |
337 | ||
a5f3027c JA |
338 | static inline int os_trim(int fd, unsigned long long start, |
339 | unsigned long long len) | |
340 | { | |
341 | uint64_t range[2]; | |
342 | ||
343 | range[0] = start; | |
344 | range[1] = len; | |
345 | ||
346 | if (!ioctl(fd, BLKDISCARD, range)) | |
347 | return 0; | |
348 | ||
349 | return errno; | |
350 | } | |
351 | ||
ebac4655 | 352 | #endif |