ceph/src/seastar/include/seastar/core/linux-aio.hh

   1 /*
   2  * This file is open source software, licensed to you under the terms
   3  * of the Apache License, Version 2.0 (the "License").  See the NOTICE file
   4  * distributed with this work for additional information regarding copyright
   5  * ownership.  You may not use this file except in compliance with the License.
   6  *
   7  * You may obtain a copy of the License at
   8  *
   9  *   http://www.apache.org/licenses/LICENSE-2.0
  10  *
  11  * Unless required by applicable law or agreed to in writing,
  12  * software distributed under the License is distributed on an
  13  * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
  14  * KIND, either express or implied.  See the License for the
  15  * specific language governing permissions and limitations
  16  * under the License.
  17  */
  18 /*
  19  * Copyright (C) 2017 ScyllaDB
  20  */
  21
  22 #pragma once
  23
  24 #include <endian.h>
  25 #include <sys/time.h>
  26 #include <sys/uio.h>
  27 #include <signal.h>
  28 #include <cstdint>
  29
  30 namespace seastar {
  31
  32 namespace internal {
  33
  34 namespace linux_abi {
  35
  36 using aio_context_t = unsigned long;
  37
  38 enum class iocb_cmd : uint16_t {
  39     PREAD = 0,
  40     PWRITE = 1,
  41     FSYNC = 2,
  42     FDSYNC = 3,
  43     POLL = 5,
  44     NOOP = 6,
  45     PREADV = 7,
  46     PWRITEV = 8,
  47 };
  48
  49 struct io_event {
  50     uint64_t data;
  51     uint64_t obj;
  52     int64_t res;
  53     int64_t res2;
  54 };
  55
  56 constexpr int IOCB_FLAG_RESFD = 1;
  57
  58 struct iocb {
  59         uint64_t   aio_data;
  60
  61 #if __BYTE_ORDER == __LITTLE_ENDIAN
  62         uint32_t   aio_key;
  63         int32_t aio_rw_flags;
  64 #elif __BYTE_ORDER == __BIG_ENDIAN
  65         int32_t aio_rw_flags;
  66         uint32_t   aio_key;
  67 #else
  68 #error bad byteorder
  69 #endif
  70
  71         iocb_cmd   aio_lio_opcode;
  72         int16_t   aio_reqprio;
  73         uint32_t   aio_fildes;
  74
  75         uint64_t   aio_buf;
  76         uint64_t   aio_nbytes;
  77         int64_t   aio_offset;
  78
  79         uint64_t   aio_reserved2;
  80
  81         uint32_t   aio_flags;
  82
  83         uint32_t   aio_resfd;
  84 };
  85
  86 struct aio_sigset {
  87     const sigset_t *sigmask;
  88     size_t sigsetsize;
  89 };
  90
  91 }
  92
  93 linux_abi::iocb make_read_iocb(int fd, uint64_t offset, void* buffer, size_t len);
  94 linux_abi::iocb make_write_iocb(int fd, uint64_t offset, const void* buffer, size_t len);
  95 linux_abi::iocb make_readv_iocb(int fd, uint64_t offset, const ::iovec* iov, size_t niov);
  96 linux_abi::iocb make_writev_iocb(int fd, uint64_t offset, const ::iovec* iov, size_t niov);
  97 linux_abi::iocb make_poll_iocb(int fd, uint32_t events);
  98
  99 void set_user_data(linux_abi::iocb& iocb, void* data);
 100 void set_nowait(linux_abi::iocb& iocb, bool nowait);
 101
 102 void set_eventfd_notification(linux_abi::iocb& iocb, int eventfd);
 103
 104 linux_abi::iocb* get_iocb(const linux_abi::io_event& ioev);
 105
 106 int io_setup(int nr_events, linux_abi::aio_context_t* io_context);
 107 int io_destroy(linux_abi::aio_context_t io_context) noexcept;
 108 int io_submit(linux_abi::aio_context_t io_context, long nr, linux_abi::iocb** iocbs);
 109 int io_cancel(linux_abi::aio_context_t io_context, linux_abi::iocb* iocb, linux_abi::io_event* result);
 110 int io_getevents(linux_abi::aio_context_t io_context, long min_nr, long nr, linux_abi::io_event* events, const ::timespec* timeout,
 111         bool force_syscall = false);
 112 int io_pgetevents(linux_abi::aio_context_t io_context, long min_nr, long nr, linux_abi::io_event* events, const ::timespec* timeout, const sigset_t* sigmask,
 113         bool force_syscall = false);
 114
 115 void setup_aio_context(size_t nr, linux_abi::aio_context_t* io_context);
 116
 117 }
 118
 119 extern bool aio_nowait_supported;
 120
 121 namespace internal {
 122
 123 inline
 124 linux_abi::iocb
 125 make_read_iocb(int fd, uint64_t offset, void* buffer, size_t len) {
 126     linux_abi::iocb iocb{};
 127     iocb.aio_lio_opcode = linux_abi::iocb_cmd::PREAD;
 128     iocb.aio_fildes = fd;
 129     iocb.aio_offset = offset;
 130     iocb.aio_buf = reinterpret_cast<uintptr_t>(buffer);
 131     iocb.aio_nbytes = len;
 132     return iocb;
 133 }
 134
 135 inline
 136 linux_abi::iocb
 137 make_write_iocb(int fd, uint64_t offset, const void* buffer, size_t len) {
 138     linux_abi::iocb iocb{};
 139     iocb.aio_lio_opcode = linux_abi::iocb_cmd::PWRITE;
 140     iocb.aio_fildes = fd;
 141     iocb.aio_offset = offset;
 142     iocb.aio_buf = reinterpret_cast<uintptr_t>(buffer);
 143     iocb.aio_nbytes = len;
 144     return iocb;
 145 }
 146
 147 inline
 148 linux_abi::iocb
 149 make_readv_iocb(int fd, uint64_t offset, const ::iovec* iov, size_t niov) {
 150     linux_abi::iocb iocb{};
 151     iocb.aio_lio_opcode = linux_abi::iocb_cmd::PREADV;
 152     iocb.aio_fildes = fd;
 153     iocb.aio_offset = offset;
 154     iocb.aio_buf = reinterpret_cast<uintptr_t>(iov);
 155     iocb.aio_nbytes = niov;
 156     return iocb;
 157 }
 158
 159 inline
 160 linux_abi::iocb
 161 make_writev_iocb(int fd, uint64_t offset, const ::iovec* iov, size_t niov) {
 162     linux_abi::iocb iocb{};
 163     iocb.aio_lio_opcode = linux_abi::iocb_cmd::PWRITEV;
 164     iocb.aio_fildes = fd;
 165     iocb.aio_offset = offset;
 166     iocb.aio_buf = reinterpret_cast<uintptr_t>(iov);
 167     iocb.aio_nbytes = niov;
 168     return iocb;
 169 }
 170
 171 inline
 172 linux_abi::iocb
 173 make_poll_iocb(int fd, uint32_t events) {
 174     linux_abi::iocb iocb{};
 175     iocb.aio_lio_opcode = linux_abi::iocb_cmd::POLL;
 176     iocb.aio_fildes = fd;
 177     iocb.aio_buf = events;
 178     return iocb;
 179 }
 180
 181 inline
 182 linux_abi::iocb
 183 make_fdsync_iocb(int fd) {
 184     linux_abi::iocb iocb{};
 185     iocb.aio_lio_opcode = linux_abi::iocb_cmd::FDSYNC;
 186     iocb.aio_fildes = fd;
 187     return iocb;
 188 }
 189
 190 inline
 191 void
 192 set_user_data(linux_abi::iocb& iocb, void* data) {
 193     iocb.aio_data = reinterpret_cast<uintptr_t>(data);
 194 }
 195
 196 template <typename T>
 197 inline T* get_user_data(const linux_abi::iocb& iocb) noexcept {
 198     return reinterpret_cast<T*>(uintptr_t(iocb.aio_data));
 199 }
 200
 201 template <typename T>
 202 inline T* get_user_data(const linux_abi::io_event& ev) noexcept {
 203     return reinterpret_cast<T*>(uintptr_t(ev.data));
 204 }
 205
 206 inline
 207 void
 208 set_eventfd_notification(linux_abi::iocb& iocb, int eventfd) {
 209     iocb.aio_flags |= linux_abi::IOCB_FLAG_RESFD;
 210     iocb.aio_resfd = eventfd;
 211 }
 212
 213 inline
 214 linux_abi::iocb*
 215 get_iocb(const linux_abi::io_event& ev) {
 216     return reinterpret_cast<linux_abi::iocb*>(uintptr_t(ev.obj));
 217 }
 218
 219 inline
 220 void
 221 set_nowait(linux_abi::iocb& iocb, bool nowait) {
 222 #ifdef RWF_NOWAIT
 223     if (aio_nowait_supported) {
 224         if (nowait) {
 225             iocb.aio_rw_flags |= RWF_NOWAIT;
 226         } else {
 227             iocb.aio_rw_flags &= ~RWF_NOWAIT;
 228         }
 229     }
 230 #endif
 231 }
 232
 233 }
 234
 235
 236 }
 237