Fix for crash with more than ~500 jobs
[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
b6959b55 48
804bb9a1
JA
49/*
50 * Can only enable this for newer glibcs, or the header and defines are
51 * missing
52 */
52ad4b94
JA
53#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 6
54#define FIO_HAVE_FALLOCATE
55#endif
78b269e3 56#if __GLIBC__ >= 2 && __GLIBC_MINOR__ >= 8
804bb9a1
JA
57#define FIO_HAVE_LINUX_FALLOCATE
58#endif
59
093729f7
JA
60#ifdef SYNC_FILE_RANGE_WAIT_BEFORE
61#define FIO_HAVE_SYNC_FILE_RANGE
62#endif
63
dc873b6f 64#define OS_MAP_ANON MAP_ANONYMOUS
ebac4655 65
4d8947de
SL
66#ifndef CLOCK_MONOTONIC
67#define CLOCK_MONOTONIC 1
68#endif
69
ebac4655 70typedef cpu_set_t os_cpu_mask_t;
4e78e405 71
6dfd46b9 72typedef struct drand48_data os_random_state_t;
ebac4655
JA
73
74/*
75 * we want fadvise64 really, but it's so tangled... later
76 */
4d8947de 77#ifdef FIO_HAVE_FADVISE
ebac4655
JA
78#define fadvise(fd, off, len, advice) \
79 posix_fadvise((fd), (off_t)(off), (len), (advice))
4d8947de 80#endif
ebac4655 81
c8f025f8
JA
82/*
83 * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want
84 * the affinity helpers to work.
85 */
86#ifndef GLIBC_2_3_2
e8462bd8
JA
87#define fio_setaffinity(pid, cpumask) \
88 sched_setaffinity((pid), sizeof(cpumask), &(cpumask))
ebac4655
JA
89#define fio_getaffinity(pid, ptr) \
90 sched_getaffinity((pid), sizeof(cpu_set_t), (ptr))
c8f025f8 91#else
e8462bd8
JA
92#define fio_setaffinity(pid, cpumask) \
93 sched_setaffinity((pid), &(cpumask))
c8f025f8
JA
94#define fio_getaffinity(pid, ptr) \
95 sched_getaffinity((pid), (ptr))
96#endif
be4ecfdf 97
1f4c0a4f
JA
98#define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask))
99#define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask))
d2ce18b5
JA
100
101static inline int fio_cpuset_init(os_cpu_mask_t *mask)
102{
103 CPU_ZERO(mask);
104 return 0;
105}
106
107static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
108{
109 return 0;
110}
6d459ee7
JA
111
112#define FIO_MAX_CPUS CPU_SETSIZE
4d8947de 113
ebac4655
JA
114static inline int ioprio_set(int which, int who, int ioprio)
115{
116 return syscall(__NR_ioprio_set, which, who, ioprio);
117}
118
47f767c1
JA
119static inline int gettid(void)
120{
121 return syscall(__NR_gettid);
122}
123
97fbdfa1
JA
124/*
125 * Just check for SPLICE_F_MOVE, if that isn't there, assume the others
126 * aren't either.
127 */
128#ifndef SPLICE_F_MOVE
129#define SPLICE_F_MOVE (0x01) /* move pages instead of copying */
130#define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */
131 /* we may still block on the fd we splice */
132 /* from/to, of course */
133#define SPLICE_F_MORE (0x04) /* expect more data */
134#define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */
8756e4d4
JA
135
136static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out,
495ee9b6 137 size_t len, unsigned int flags)
8756e4d4 138{
97fbdfa1 139 return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags);
8756e4d4
JA
140}
141
142static inline int tee(int fdin, int fdout, size_t len, unsigned int flags)
143{
97fbdfa1 144 return syscall(__NR_sys_tee, fdin, fdout, len, flags);
8756e4d4
JA
145}
146
147static inline int vmsplice(int fd, const struct iovec *iov,
148 unsigned long nr_segs, unsigned int flags)
149{
97fbdfa1 150 return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags);
8756e4d4 151}
97fbdfa1 152#endif
8756e4d4 153
3feedc60
JA
154#define SPLICE_DEF_SIZE (64*1024)
155
b6959b55 156#ifdef FIO_HAVE_SYSLET
bf0dc8fa
IM
157
158struct syslet_uatom;
159struct async_head_user;
160
a4f4fdd7
JA
161/*
162 * syslet stuff
163 */
bf0dc8fa
IM
164static inline struct syslet_uatom *
165async_exec(struct syslet_uatom *atom, struct async_head_user *ahu)
a4f4fdd7 166{
2b52511f 167 return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu);
a4f4fdd7
JA
168}
169
bf0dc8fa
IM
170static inline long
171async_wait(unsigned long min_wait_events, unsigned long user_ring_idx,
172 struct async_head_user *ahu)
a4f4fdd7 173{
bf0dc8fa
IM
174 return syscall(__NR_async_wait, min_wait_events,
175 user_ring_idx, ahu);
a4f4fdd7
JA
176}
177
7756b0d0 178static inline long async_thread(void *event, struct async_head_user *ahu)
a4f4fdd7 179{
7756b0d0 180 return syscall(__NR_async_thread, event, ahu);
a4f4fdd7
JA
181}
182
183static inline long umem_add(unsigned long *uptr, unsigned long inc)
184{
185 return syscall(__NR_umem_add, uptr, inc);
186}
b6959b55 187#endif /* FIO_HAVE_SYSLET */
a4f4fdd7 188
6cefbe33
JA
189enum {
190 IOPRIO_CLASS_NONE,
191 IOPRIO_CLASS_RT,
192 IOPRIO_CLASS_BE,
193 IOPRIO_CLASS_IDLE,
194};
195
ebac4655
JA
196enum {
197 IOPRIO_WHO_PROCESS = 1,
198 IOPRIO_WHO_PGRP,
199 IOPRIO_WHO_USER,
200};
201
6cefbe33 202#define IOPRIO_BITS 16
ebac4655
JA
203#define IOPRIO_CLASS_SHIFT 13
204
205#ifndef BLKGETSIZE64
206#define BLKGETSIZE64 _IOR(0x12,114,size_t)
207#endif
208
e5b401d4
JA
209#ifndef BLKFLSBUF
210#define BLKFLSBUF _IO(0x12,97)
211#endif
212
a5f3027c
JA
213#ifndef BLKDISCARD
214#define BLKDISCARD _IO(0x12,119)
215#endif
216
e2e58886 217static inline int blockdev_invalidate_cache(struct fio_file *f)
e5b401d4 218{
ecc314ba 219 return ioctl(f->fd, BLKFLSBUF);
e5b401d4
JA
220}
221
ecc314ba 222static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes)
ebac4655 223{
ecc314ba 224 if (!ioctl(f->fd, BLKGETSIZE64, bytes))
ebac4655
JA
225 return 0;
226
227 return errno;
228}
229
32cd46a0
JA
230static inline unsigned long long os_phys_mem(void)
231{
232 long pagesize, pages;
233
234 pagesize = sysconf(_SC_PAGESIZE);
235 pages = sysconf(_SC_PHYS_PAGES);
236 if (pages == -1 || pagesize == -1)
237 return 0;
238
239 return (unsigned long long) pages * (unsigned long long) pagesize;
240}
241
6dfd46b9
JA
242static inline void os_random_seed(unsigned long seed, os_random_state_t *rs)
243{
244 srand48_r(seed, rs);
245}
246
247static inline long os_random_long(os_random_state_t *rs)
248{
249 long val;
250
251 lrand48_r(rs, &val);
252 return val;
253}
254
8cc7afa9 255static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev)
07e5b264
JA
256{
257 struct raw_config_request rq;
258 int fd;
259
260 if (major(dev) != RAW_MAJOR)
8cc7afa9 261 return 1;
07e5b264
JA
262
263 /*
264 * we should be able to find /dev/rawctl or /dev/raw/rawctl
265 */
266 fd = open("/dev/rawctl", O_RDONLY);
267 if (fd < 0) {
268 fd = open("/dev/raw/rawctl", O_RDONLY);
269 if (fd < 0)
8cc7afa9 270 return 1;
07e5b264
JA
271 }
272
273 rq.raw_minor = minor(dev);
274 if (ioctl(fd, RAW_GETBIND, &rq) < 0) {
275 close(fd);
8cc7afa9 276 return 1;
07e5b264
JA
277 }
278
279 close(fd);
280 *majdev = rq.block_major;
281 *mindev = rq.block_minor;
8cc7afa9 282 return 0;
07e5b264
JA
283}
284
5b6f5c66 285#ifdef O_NOATIME
5921e80c 286#define FIO_O_NOATIME O_NOATIME
5b6f5c66
JA
287#else
288#define FIO_O_NOATIME 0
289#endif
5921e80c 290
a1c58075
JA
291#ifdef MADV_REMOVE
292#define FIO_MADV_FREE MADV_REMOVE
293#endif
294
610a730c
JA
295#if __BYTE_ORDER == __LITTLE_ENDIAN
296#define FIO_LITTLE_ENDIAN
297#elif __BYTE_ORDER == __BIG_ENDIAN
298#define FIO_BIG_ENDIAN
299#else
300#error "Unknown endianness"
301#endif
302
ff245192
JA
303#define fio_swap16(x) __bswap_16(x)
304#define fio_swap32(x) __bswap_32(x)
305#define fio_swap64(x) __bswap_64(x)
306
eb7ccf38
JA
307#define CACHE_LINE_FILE \
308 "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size"
309
310static inline int arch_cache_line_size(void)
311{
312 char size[32];
313 int fd, ret;
314
315 fd = open(CACHE_LINE_FILE, O_RDONLY);
316 if (fd < 0)
317 return -1;
318
319 ret = read(fd, size, sizeof(size));
ec76f15a
JA
320
321 close(fd);
322
eb7ccf38
JA
323 if (ret <= 0)
324 return -1;
ec76f15a
JA
325 else
326 return atoi(size);
eb7ccf38
JA
327}
328
2e3bd4c2
JA
329static inline unsigned long long get_fs_size(const char *path)
330{
331 unsigned long long ret;
332 struct statfs s;
333
334 if (statfs(path, &s) < 0)
335 return -1ULL;
336
337 ret = s.f_bsize;
338 ret *= (unsigned long long) s.f_bfree;
339 return ret;
340}
341
a5f3027c
JA
342static inline int os_trim(int fd, unsigned long long start,
343 unsigned long long len)
344{
345 uint64_t range[2];
346
347 range[0] = start;
348 range[1] = len;
349
350 if (!ioctl(fd, BLKDISCARD, range))
351 return 0;
352
353 return errno;
354}
355
ebac4655 356#endif