linux文件拷贝-sendfile

SENDFILE(2)                Linux Programmer's Manual               SENDFILE(2)

NAME
       sendfile - transfer data between file descriptors  在两个文件描述符之间传输数据

SYNOPSIS
       #include <sys/sendfile.h>

       ssize_t sendfile(int out_fd, int in_fd, off_t *offset, size_t count);

DESCRIPTION
       sendfile()  copies  data  between  one  file  descriptor  and  another.
       Because this copying is done within  the  kernel,  sendfile()  is  more
       efficient  than  the  combination  of read(2) and write(2), which would
       require transferring data to and from user space.

拷贝文件，因为此拷贝过程在内核态完成，所以效率比使用读写函数更好，读取函数通过用户空间传输数据。

       in_fd should be a file descriptor opened for reading and out_fd  should
       be a descriptor opened for writing.

in_fd指定以读打开的文件，out_fd指定以写打开。

       If  offset  is  not NULL, then it points to a variable holding the file
       offset from which sendfile() will start reading data from in_fd.   When
       sendfile() returns, this variable will be set to the offset of the byte
       following the last byte that was read.  If offset  is  not  NULL,  then
       sendfile() does not modify the file offset of in_fd; otherwise the file
       offset is adjusted to reflect the number of bytes read from in_fd.

如果offset不为空，从此偏移开始读取，返回时设置此读取后的偏移。

       If offset is NULL, then data will be read from in_fd  starting  at  the
       file offset, and the file offset will be updated by the call.

如果偏移为空，将从文件当前偏移读取，完成调用后更新文件调用。

       count is the number of bytes to copy between the file descriptors.

拷贝的字节数

       The   in_fd   argument   must  correspond  to  a  file  which  supports
       mmap(2)-like operations (i.e., it cannot be a socket).

in_fd文件描述符必须支持内存映射，不能是socket

       In Linux kernels before 2.6.33, out_fd must refer to a  socket.   Since
       Linux  2.6.33  it can be any file.  If it is a regular file, then send‐
       file() changes the file offset appropriately.

2.6.33之前out_fd必须为socket，之后可以为任何文件。如果是常规文件，sendfile必变文件描述的偏移。

RETURN VALUE
       If the transfer was successful, the number of bytes written  to  out_fd
       is returned.  Note that a successful call to sendfile() may write fewer
       bytes than requested; the caller should be prepared to retry  the  call
       if there were unsent bytes.  See also NOTES.

传输成功，字节数写入out_fd,sendfile调用成功写入的字节数可能少于请求的。如果有末发送的字节，应该重发。

       On error, -1 is returned, and errno is set appropriately.

失败返回-1，并设置errno

ERRORS
       EAGAIN Nonblocking I/O has been selected using O_NONBLOCK and the write
              would block.

       EBADF  The input file was not opened for reading or the output file was
              not opened for writing.

       EFAULT Bad address.

       EINVAL Descriptor  is not valid or locked, or an mmap(2)-like operation
              is not available for in_fd, or count is negative.

       EINVAL out_fd has the O_APPEND flag set.  This is  not  currently  sup‐
              ported by sendfile().

       EIO    Unspecified error while reading from in_fd.

       ENOMEM Insufficient memory to read from in_fd.

       EOVERFLOW
              count  is too large, the operation would result in exceeding the
              maximum size of either the input file or the output file.

       ESPIPE offset is not NULL but the input file is not seek(2)-able.

VERSIONS
       sendfile() first appeared in Linux 2.2.  The  include  file  <sys/send‐
       file.h> is present since glibc 2.1.

CONFORMING TO
       Not specified in POSIX.1-2001, nor in other standards.

       Other  UNIX  systems  implement sendfile() with different semantics and
       prototypes.  It should not be used in portable programs.

NOTES
       sendfile() will transfer  at  most  0x7ffff000  (2,147,479,552)  bytes,
       returning  the  number of bytes actually transferred.  (This is true on
       both 32-bit and 64-bit systems.)

sendfile最多传输0x7ffff000字节

       If you plan to use sendfile() for sending files to a  TCP  socket,  but
       need  to  send some header data in front of the file contents, you will
       find it useful to employ the TCP_CORK option, described in  tcp(7),  to
       minimize the number of packets and to tune performance.

如果计划使用sendfile将文件发送到TCP socket，但是需要在内容前发达一些头数据。参考TCP_CORK选项。

       In  Linux  2.4  and earlier, out_fd could also refer to a regular file;
       this possibility went away in the Linux 2.6.x kernel  series,  but  was
       restored in Linux 2.6.33.

2.6.33之后可以为普通文件

       The  original  Linux  sendfile() system call was not designed to handle
       large file offsets.  Consequently, Linux 2.4 added sendfile64(), with a
       wider type for the offset argument.  The glibc sendfile() wrapper func‐
       tion transparently deals with the kernel differences.

2.4之后，sendfile64，用来支大文件传输

       Applications may wish to fall back  to  read(2)/write(2)  in  the  case
       where sendfile() fails with EINVAL or ENOSYS.

       If  out_fd  refers  to a socket or pipe with zero-copy support, callers
       must ensure the transferred portions of the file referred to  by  in_fd
       remain  unmodified until the reader on the other end of out_fd has con‐
       sumed the transferred data.

如果out_fd是socket或pipe，零拷贝支持，调用方必需确保in_fd保持末修改的状态，直至out_fd写完

       The Linux-specific splice(2) call supports  transferring  data  between
       arbitrary file descriptors provided one (or both) of them is a pipe.

特定内核 splice调用支持在任务文件描述符之前传递数据

SEE ALSO
       copy_file_range(2), mmap(2), open(2), socket(2), splice(2)

COLOPHON
       This  page  is  part of release 4.15 of the Linux man-pages project.  A
       description of the project, information about reporting bugs,  and  the
       latest     version     of     this    page,    can    be    found    at
       https://www.kernel.org/doc/man-pages/.

Linux                             2017-09-15                       SENDFILE(2)