| 1 | #ifndef FIO_OS_LINUX_H |
| 2 | #define FIO_OS_LINUX_H |
| 3 | |
| 4 | #define FIO_OS os_linux |
| 5 | |
| 6 | #include <sys/ioctl.h> |
| 7 | #include <sys/uio.h> |
| 8 | #include <sys/syscall.h> |
| 9 | #include <sys/vfs.h> |
| 10 | #include <sys/mman.h> |
| 11 | #include <unistd.h> |
| 12 | #include <fcntl.h> |
| 13 | #include <errno.h> |
| 14 | #include <sched.h> |
| 15 | #include <linux/unistd.h> |
| 16 | #include <linux/raw.h> |
| 17 | #include <linux/major.h> |
| 18 | |
| 19 | #include "binject.h" |
| 20 | #include "../file.h" |
| 21 | |
| 22 | #define FIO_HAVE_CPU_AFFINITY |
| 23 | #define FIO_HAVE_DISK_UTIL |
| 24 | #define FIO_HAVE_SGIO |
| 25 | #define FIO_HAVE_IOPRIO |
| 26 | #define FIO_HAVE_IOSCHED_SWITCH |
| 27 | #define FIO_HAVE_ODIRECT |
| 28 | #define FIO_HAVE_HUGETLB |
| 29 | #define FIO_HAVE_RAWBIND |
| 30 | #define FIO_HAVE_BLKTRACE |
| 31 | #define FIO_HAVE_PSHARED_MUTEX |
| 32 | #define FIO_HAVE_CL_SIZE |
| 33 | #define FIO_HAVE_CGROUPS |
| 34 | #define FIO_HAVE_FS_STAT |
| 35 | #define FIO_HAVE_TRIM |
| 36 | #define FIO_HAVE_BINJECT |
| 37 | #define FIO_HAVE_GETTID |
| 38 | #define FIO_USE_GENERIC_INIT_RANDOM_STATE |
| 39 | |
| 40 | #ifdef MAP_HUGETLB |
| 41 | #define FIO_HAVE_MMAP_HUGE |
| 42 | #endif |
| 43 | |
| 44 | #define OS_MAP_ANON MAP_ANONYMOUS |
| 45 | |
| 46 | typedef cpu_set_t os_cpu_mask_t; |
| 47 | |
| 48 | typedef struct drand48_data os_random_state_t; |
| 49 | |
| 50 | #ifdef CONFIG_3ARG_AFFINITY |
| 51 | #define fio_setaffinity(pid, cpumask) \ |
| 52 | sched_setaffinity((pid), sizeof(cpumask), &(cpumask)) |
| 53 | #define fio_getaffinity(pid, ptr) \ |
| 54 | sched_getaffinity((pid), sizeof(cpu_set_t), (ptr)) |
| 55 | #elif defined(CONFIG_2ARG_AFFINITY) |
| 56 | #define fio_setaffinity(pid, cpumask) \ |
| 57 | sched_setaffinity((pid), &(cpumask)) |
| 58 | #define fio_getaffinity(pid, ptr) \ |
| 59 | sched_getaffinity((pid), (ptr)) |
| 60 | #endif |
| 61 | |
| 62 | #define fio_cpu_clear(mask, cpu) (void) CPU_CLR((cpu), (mask)) |
| 63 | #define fio_cpu_set(mask, cpu) (void) CPU_SET((cpu), (mask)) |
| 64 | #define fio_cpu_isset(mask, cpu) CPU_ISSET((cpu), (mask)) |
| 65 | #define fio_cpu_count(maks) CPU_COUNT((mask)) |
| 66 | |
| 67 | static inline int fio_cpuset_init(os_cpu_mask_t *mask) |
| 68 | { |
| 69 | CPU_ZERO(mask); |
| 70 | return 0; |
| 71 | } |
| 72 | |
| 73 | static inline int fio_cpuset_exit(os_cpu_mask_t *mask) |
| 74 | { |
| 75 | return 0; |
| 76 | } |
| 77 | |
| 78 | #define FIO_MAX_CPUS CPU_SETSIZE |
| 79 | |
| 80 | enum { |
| 81 | IOPRIO_CLASS_NONE, |
| 82 | IOPRIO_CLASS_RT, |
| 83 | IOPRIO_CLASS_BE, |
| 84 | IOPRIO_CLASS_IDLE, |
| 85 | }; |
| 86 | |
| 87 | enum { |
| 88 | IOPRIO_WHO_PROCESS = 1, |
| 89 | IOPRIO_WHO_PGRP, |
| 90 | IOPRIO_WHO_USER, |
| 91 | }; |
| 92 | |
| 93 | #define IOPRIO_BITS 16 |
| 94 | #define IOPRIO_CLASS_SHIFT 13 |
| 95 | |
| 96 | static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio) |
| 97 | { |
| 98 | /* |
| 99 | * If no class is set, assume BE |
| 100 | */ |
| 101 | if (!ioprio_class) |
| 102 | ioprio_class = IOPRIO_CLASS_BE; |
| 103 | |
| 104 | ioprio |= ioprio_class << IOPRIO_CLASS_SHIFT; |
| 105 | return syscall(__NR_ioprio_set, which, who, ioprio); |
| 106 | } |
| 107 | |
| 108 | static inline int gettid(void) |
| 109 | { |
| 110 | return syscall(__NR_gettid); |
| 111 | } |
| 112 | |
| 113 | #define SPLICE_DEF_SIZE (64*1024) |
| 114 | |
| 115 | #ifndef BLKGETSIZE64 |
| 116 | #define BLKGETSIZE64 _IOR(0x12,114,size_t) |
| 117 | #endif |
| 118 | |
| 119 | #ifndef BLKFLSBUF |
| 120 | #define BLKFLSBUF _IO(0x12,97) |
| 121 | #endif |
| 122 | |
| 123 | #ifndef BLKDISCARD |
| 124 | #define BLKDISCARD _IO(0x12,119) |
| 125 | #endif |
| 126 | |
| 127 | static inline int blockdev_invalidate_cache(struct fio_file *f) |
| 128 | { |
| 129 | return ioctl(f->fd, BLKFLSBUF); |
| 130 | } |
| 131 | |
| 132 | static inline int blockdev_size(struct fio_file *f, unsigned long long *bytes) |
| 133 | { |
| 134 | if (!ioctl(f->fd, BLKGETSIZE64, bytes)) |
| 135 | return 0; |
| 136 | |
| 137 | return errno; |
| 138 | } |
| 139 | |
| 140 | static inline unsigned long long os_phys_mem(void) |
| 141 | { |
| 142 | long pagesize, pages; |
| 143 | |
| 144 | pagesize = sysconf(_SC_PAGESIZE); |
| 145 | pages = sysconf(_SC_PHYS_PAGES); |
| 146 | if (pages == -1 || pagesize == -1) |
| 147 | return 0; |
| 148 | |
| 149 | return (unsigned long long) pages * (unsigned long long) pagesize; |
| 150 | } |
| 151 | |
| 152 | static inline void os_random_seed(unsigned long seed, os_random_state_t *rs) |
| 153 | { |
| 154 | srand48_r(seed, rs); |
| 155 | } |
| 156 | |
| 157 | static inline long os_random_long(os_random_state_t *rs) |
| 158 | { |
| 159 | long val; |
| 160 | |
| 161 | lrand48_r(rs, &val); |
| 162 | return val; |
| 163 | } |
| 164 | |
| 165 | static inline int fio_lookup_raw(dev_t dev, int *majdev, int *mindev) |
| 166 | { |
| 167 | struct raw_config_request rq; |
| 168 | int fd; |
| 169 | |
| 170 | if (major(dev) != RAW_MAJOR) |
| 171 | return 1; |
| 172 | |
| 173 | /* |
| 174 | * we should be able to find /dev/rawctl or /dev/raw/rawctl |
| 175 | */ |
| 176 | fd = open("/dev/rawctl", O_RDONLY); |
| 177 | if (fd < 0) { |
| 178 | fd = open("/dev/raw/rawctl", O_RDONLY); |
| 179 | if (fd < 0) |
| 180 | return 1; |
| 181 | } |
| 182 | |
| 183 | rq.raw_minor = minor(dev); |
| 184 | if (ioctl(fd, RAW_GETBIND, &rq) < 0) { |
| 185 | close(fd); |
| 186 | return 1; |
| 187 | } |
| 188 | |
| 189 | close(fd); |
| 190 | *majdev = rq.block_major; |
| 191 | *mindev = rq.block_minor; |
| 192 | return 0; |
| 193 | } |
| 194 | |
| 195 | #ifdef O_NOATIME |
| 196 | #define FIO_O_NOATIME O_NOATIME |
| 197 | #else |
| 198 | #define FIO_O_NOATIME 0 |
| 199 | #endif |
| 200 | |
| 201 | #ifdef O_ATOMIC |
| 202 | #define OS_O_ATOMIC O_ATOMIC |
| 203 | #else |
| 204 | #define OS_O_ATOMIC 040000000 |
| 205 | #endif |
| 206 | |
| 207 | #ifdef MADV_REMOVE |
| 208 | #define FIO_MADV_FREE MADV_REMOVE |
| 209 | #endif |
| 210 | |
| 211 | #define fio_swap16(x) __bswap_16(x) |
| 212 | #define fio_swap32(x) __bswap_32(x) |
| 213 | #define fio_swap64(x) __bswap_64(x) |
| 214 | |
| 215 | #define CACHE_LINE_FILE \ |
| 216 | "/sys/devices/system/cpu/cpu0/cache/index0/coherency_line_size" |
| 217 | |
| 218 | static inline int arch_cache_line_size(void) |
| 219 | { |
| 220 | char size[32]; |
| 221 | int fd, ret; |
| 222 | |
| 223 | fd = open(CACHE_LINE_FILE, O_RDONLY); |
| 224 | if (fd < 0) |
| 225 | return -1; |
| 226 | |
| 227 | ret = read(fd, size, sizeof(size)); |
| 228 | |
| 229 | close(fd); |
| 230 | |
| 231 | if (ret <= 0) |
| 232 | return -1; |
| 233 | else |
| 234 | return atoi(size); |
| 235 | } |
| 236 | |
| 237 | static inline unsigned long long get_fs_size(const char *path) |
| 238 | { |
| 239 | unsigned long long ret; |
| 240 | struct statfs s; |
| 241 | |
| 242 | if (statfs(path, &s) < 0) |
| 243 | return -1ULL; |
| 244 | |
| 245 | ret = s.f_bsize; |
| 246 | ret *= (unsigned long long) s.f_bfree; |
| 247 | return ret; |
| 248 | } |
| 249 | |
| 250 | static inline int os_trim(int fd, unsigned long long start, |
| 251 | unsigned long long len) |
| 252 | { |
| 253 | uint64_t range[2]; |
| 254 | |
| 255 | range[0] = start; |
| 256 | range[1] = len; |
| 257 | |
| 258 | if (!ioctl(fd, BLKDISCARD, range)) |
| 259 | return 0; |
| 260 | |
| 261 | return errno; |
| 262 | } |
| 263 | |
| 264 | #ifdef CONFIG_SCHED_IDLE |
| 265 | static inline int fio_set_sched_idle(void) |
| 266 | { |
| 267 | struct sched_param p = { .sched_priority = 0, }; |
| 268 | return sched_setscheduler(gettid(), SCHED_IDLE, &p); |
| 269 | } |
| 270 | #endif |
| 271 | |
| 272 | #endif |