| 1 | #ifndef FIO_OS_LINUX_H |
| 2 | #define FIO_OS_LINUX_H |
| 3 | |
| 4 | #include <sys/ioctl.h> |
| 5 | #include <sys/uio.h> |
| 6 | #include <sys/syscall.h> |
| 7 | #include <unistd.h> |
| 8 | #include <fcntl.h> |
| 9 | #include <linux/unistd.h> |
| 10 | #include <linux/raw.h> |
| 11 | #include <linux/major.h> |
| 12 | |
| 13 | #include "indirect.h" |
| 14 | |
| 15 | #define FIO_HAVE_LIBAIO |
| 16 | #define FIO_HAVE_POSIXAIO |
| 17 | #define FIO_HAVE_FADVISE |
| 18 | #define FIO_HAVE_CPU_AFFINITY |
| 19 | #define FIO_HAVE_DISK_UTIL |
| 20 | #define FIO_HAVE_SGIO |
| 21 | #define FIO_HAVE_IOPRIO |
| 22 | #define FIO_HAVE_SPLICE |
| 23 | #define FIO_HAVE_IOSCHED_SWITCH |
| 24 | #define FIO_HAVE_ODIRECT |
| 25 | #define FIO_HAVE_HUGETLB |
| 26 | #define FIO_HAVE_RAWBIND |
| 27 | #define FIO_HAVE_BLKTRACE |
| 28 | #define FIO_HAVE_STRSEP |
| 29 | #define FIO_HAVE_FALLOCATE |
| 30 | #define FIO_HAVE_POSIXAIO_FSYNC |
| 31 | |
| 32 | #define OS_MAP_ANON MAP_ANONYMOUS |
| 33 | |
| 34 | #ifndef CLOCK_MONOTONIC |
| 35 | #define CLOCK_MONOTONIC 1 |
| 36 | #endif |
| 37 | |
| 38 | #ifdef FIO_HAVE_CPU_AFFINITY |
| 39 | typedef cpu_set_t os_cpu_mask_t; |
| 40 | #else |
| 41 | typedef int os_cpu_mask_t; |
| 42 | #endif |
| 43 | typedef struct drand48_data os_random_state_t; |
| 44 | |
| 45 | /* |
| 46 | * we want fadvise64 really, but it's so tangled... later |
| 47 | */ |
| 48 | #ifdef FIO_HAVE_FADVISE |
| 49 | #define fadvise(fd, off, len, advice) \ |
| 50 | posix_fadvise((fd), (off_t)(off), (len), (advice)) |
| 51 | #endif |
| 52 | |
| 53 | /* |
| 54 | * If you are on an ancient glibc (2.3.2), then define GLIBC_2_3_2 if you want |
| 55 | * the affinity helpers to work. |
| 56 | */ |
| 57 | #ifdef FIO_HAVE_CPU_AFFINITY |
| 58 | #ifndef GLIBC_2_3_2 |
| 59 | #define fio_setaffinity(td) \ |
| 60 | sched_setaffinity((td)->pid, sizeof((td)->o.cpumask), &(td)->o.cpumask) |
| 61 | #define fio_getaffinity(pid, ptr) \ |
| 62 | sched_getaffinity((pid), sizeof(cpu_set_t), (ptr)) |
| 63 | #else |
| 64 | #define fio_setaffinity(td) \ |
| 65 | sched_setaffinity((td)->pid, &(td)->o.cpumask) |
| 66 | #define fio_getaffinity(pid, ptr) \ |
| 67 | sched_getaffinity((pid), (ptr)) |
| 68 | #endif |
| 69 | #endif |
| 70 | |
| 71 | static inline int ioprio_set(int which, int who, int ioprio) |
| 72 | { |
| 73 | return syscall(__NR_ioprio_set, which, who, ioprio); |
| 74 | } |
| 75 | |
| 76 | /* |
| 77 | * Just check for SPLICE_F_MOVE, if that isn't there, assume the others |
| 78 | * aren't either. |
| 79 | */ |
| 80 | #ifndef SPLICE_F_MOVE |
| 81 | #define SPLICE_F_MOVE (0x01) /* move pages instead of copying */ |
| 82 | #define SPLICE_F_NONBLOCK (0x02) /* don't block on the pipe splicing (but */ |
| 83 | /* we may still block on the fd we splice */ |
| 84 | /* from/to, of course */ |
| 85 | #define SPLICE_F_MORE (0x04) /* expect more data */ |
| 86 | #define SPLICE_F_GIFT (0x08) /* pages passed in are a gift */ |
| 87 | |
| 88 | static inline int splice(int fdin, loff_t *off_in, int fdout, loff_t *off_out, |
| 89 | size_t len, unsigned int flags) |
| 90 | { |
| 91 | return syscall(__NR_sys_splice, fdin, off_in, fdout, off_out, len, flags); |
| 92 | } |
| 93 | |
| 94 | static inline int tee(int fdin, int fdout, size_t len, unsigned int flags) |
| 95 | { |
| 96 | return syscall(__NR_sys_tee, fdin, fdout, len, flags); |
| 97 | } |
| 98 | |
| 99 | static inline int vmsplice(int fd, const struct iovec *iov, |
| 100 | unsigned long nr_segs, unsigned int flags) |
| 101 | { |
| 102 | return syscall(__NR_sys_vmsplice, fd, iov, nr_segs, flags); |
| 103 | } |
| 104 | #endif |
| 105 | |
| 106 | #define SPLICE_DEF_SIZE (64*1024) |
| 107 | |
| 108 | #ifdef FIO_HAVE_SYSLET |
| 109 | |
| 110 | struct syslet_uatom; |
| 111 | struct async_head_user; |
| 112 | |
| 113 | /* |
| 114 | * syslet stuff |
| 115 | */ |
| 116 | static inline struct syslet_uatom * |
| 117 | async_exec(struct syslet_uatom *atom, struct async_head_user *ahu) |
| 118 | { |
| 119 | return (struct syslet_uatom *) syscall(__NR_async_exec, atom, ahu); |
| 120 | } |
| 121 | |
| 122 | static inline long |
| 123 | async_wait(unsigned long min_wait_events, unsigned long user_ring_idx, |
| 124 | struct async_head_user *ahu) |
| 125 | { |
| 126 | return syscall(__NR_async_wait, min_wait_events, |
| 127 | user_ring_idx, ahu); |
| 128 | } |
| 129 | |
| 130 | static inline long async_thread(void *event, struct async_head_user *ahu) |
| 131 | { |
| 132 | return syscall(__NR_async_thread, event, ahu); |
| 133 | } |
| 134 | |
| 135 | static inline long umem_add(unsigned long *uptr, unsigned long inc) |
| 136 | { |
| 137 | return syscall(__NR_umem_add, uptr, inc); |
| 138 | } |
| 139 | #endif /* FIO_HAVE_SYSLET */ |
| 140 | |
| 141 | enum { |
| 142 | IOPRIO_CLASS_NONE, |
| 143 | IOPRIO_CLASS_RT, |
| 144 | IOPRIO_CLASS_BE, |
| 145 | IOPRIO_CLASS_IDLE, |
| 146 | }; |
| 147 | |
| 148 | enum { |
| 149 | IOPRIO_WHO_PROCESS = 1, |
| 150 | IOPRIO_WHO_PGRP, |
| 151 | IOPRIO_WHO_USER, |
| 152 | }; |
| 153 | |
| 154 | #define IOPRIO_BITS 16 |
| 155 | #define IOPRIO_CLASS_SHIFT 13 |
| 156 | |
| 157 | #ifndef BLKGETSIZE64 |
| 158 | #define BLKGETSIZE64 _IOR(0x12,114,size_t) |
| 159 | #endif |
| 160 | |
| 161 | #ifndef BLKFLSBUF |
| 162 | #define BLKFLSBUF _IO(0x12,97) |
| 163 | #endif |
| 164 | |
| 165 | static inline int blockdev_invalidate_cache(int fd) |
| 166 | { |
| 167 | return ioctl(fd, BLKFLSBUF); |
| 168 | } |
| 169 | |
| 170 | static inline int blockdev_size(int fd, unsigned long long *bytes) |
| 171 | { |
| 172 | if (!ioctl(fd, BLKGETSIZE64, bytes)) |
| 173 | return 0; |
| 174 | |
| 175 | return errno; |
| 176 | } |
| 177 | |
| 178 | static inline unsigned long long os_phys_mem(void) |
| 179 | { |
| 180 | long pagesize, pages; |
| 181 | |
| 182 | pagesize = sysconf(_SC_PAGESIZE); |
| 183 | pages = sysconf(_SC_PHYS_PAGES); |
| 184 | if (pages == -1 || pagesize == -1) |
| 185 | return 0; |
| 186 | |
| 187 | return (unsigned long long) pages * (unsigned long long) pagesize; |
| 188 | } |
| 189 | |
| 190 | static inline void os_random_seed(unsigned long seed, os_random_state_t *rs) |
| 191 | { |
| 192 | srand48_r(seed, rs); |
| 193 | } |
| 194 | |
| 195 | static inline long os_random_long(os_random_state_t *rs) |
| 196 | { |
| 197 | long val; |
| 198 | |
| 199 | lrand48_r(rs, &val); |
| 200 | return val; |
| 201 | } |
| 202 | |
| 203 | static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev) |
| 204 | { |
| 205 | struct raw_config_request rq; |
| 206 | int fd; |
| 207 | |
| 208 | if (major(dev) != RAW_MAJOR) |
| 209 | return 1; |
| 210 | |
| 211 | /* |
| 212 | * we should be able to find /dev/rawctl or /dev/raw/rawctl |
| 213 | */ |
| 214 | fd = open("/dev/rawctl", O_RDONLY); |
| 215 | if (fd < 0) { |
| 216 | fd = open("/dev/raw/rawctl", O_RDONLY); |
| 217 | if (fd < 0) |
| 218 | return 1; |
| 219 | } |
| 220 | |
| 221 | rq.raw_minor = minor(dev); |
| 222 | if (ioctl(fd, RAW_GETBIND, &rq) < 0) { |
| 223 | close(fd); |
| 224 | return 1; |
| 225 | } |
| 226 | |
| 227 | close(fd); |
| 228 | *majdev = rq.block_major; |
| 229 | *mindev = rq.block_minor; |
| 230 | return 0; |
| 231 | } |
| 232 | |
| 233 | #ifdef O_NOATIME |
| 234 | #define FIO_O_NOATIME O_NOATIME |
| 235 | #else |
| 236 | #define FIO_O_NOATIME 0 |
| 237 | #endif |
| 238 | |
| 239 | #endif |