path: root/man2/splice.2

.\" This manpage is Copyright (C) 2006 Jens Axboe
.\" and Copyright (C) 2006 Michael Kerrisk <mtk.manpages@gmail.com>
.\"
.\" SPDX-License-Identifier: Linux-man-pages-copyleft
.\"
.TH splice 2 (date) "Linux man-pages (unreleased)"
.SH NAME
splice \- splice data to/from a pipe
.SH LIBRARY
Standard C library
.RI ( libc ", " \-lc )
.SH SYNOPSIS
.nf
.BR "#define _GNU_SOURCE" "         /* See feature_test_macros(7) */"
.B #define _FILE_OFFSET_BITS  64
.B #include <fcntl.h>
.P
.BI "ssize_t splice(int " fd_in ", off_t *_Nullable " off_in ,
.BI "               int " fd_out ", off_t *_Nullable " off_out ,
.BI "               size_t " len ", unsigned int " flags );
.\" Return type was long before glibc 2.7
.fi
.SH DESCRIPTION
.BR splice ()
moves data between two file descriptors
without copying between kernel address space and user address space.
It transfers up to
.I len
bytes of data from the file descriptor
.I fd_in
to the file descriptor
.IR fd_out ,
where one of the file descriptors must refer to a pipe.
.P
The following semantics apply for
.I fd_in
and
.IR off_in :
.IP \[bu] 3
If
.I fd_in
refers to a pipe, then
.I off_in
must be NULL.
.IP \[bu]
If
.I fd_in
does not refer to a pipe and
.I off_in
is NULL, then bytes are read from
.I fd_in
starting from the file offset,
and the file offset is adjusted appropriately.
.IP \[bu]
If
.I fd_in
does not refer to a pipe and
.I off_in
is not NULL, then
.I off_in
must point to a buffer which specifies the starting
offset from which bytes will be read from
.IR fd_in ;
in this case, the file offset of
.I fd_in
is not changed.
.P
Analogous statements apply for
.I fd_out
and
.IR off_out .
.P
The
.I flags
argument is a bit mask that is composed by ORing together
zero or more of the following values:
.TP
.B SPLICE_F_MOVE
Attempt to move pages instead of copying.
This is only a hint to the kernel:
pages may still be copied if the kernel cannot move the
pages from the pipe, or if
the pipe buffers don't refer to full pages.
The initial implementation of this flag was buggy:
therefore starting in Linux 2.6.21 it is a no-op
(but is still permitted in a
.BR splice ()
call);
in the future, a correct implementation may be restored.
.TP
.B SPLICE_F_NONBLOCK
Do not block on I/O.
This makes the splice pipe operations nonblocking, but
.BR splice ()
may nevertheless block because the file descriptors that
are spliced to/from may block (unless they have the
.B O_NONBLOCK
flag set).
.TP
.B SPLICE_F_MORE
More data will be coming in a subsequent splice.
This is a helpful hint when
the
.I fd_out
refers to a socket (see also the description of
.B MSG_MORE
in
.BR send (2),
and the description of
.B TCP_CORK
in
.BR tcp (7)).
.TP
.B SPLICE_F_GIFT
Unused for
.BR splice ();
see
.BR vmsplice (2).
.SH RETURN VALUE
Upon successful completion,
.BR splice ()
returns the number of bytes
spliced to or from the pipe.
.P
A return value of 0 means end of input.
If
.I fd_in
refers to a pipe, then this means that there was no data to transfer,
and it would not make sense to block because there are no writers
connected to the write end of the pipe.
.P
On error,
.BR splice ()
returns \-1 and
.I errno
is set to indicate the error.
.SH ERRORS
.TP
.B EAGAIN
.B SPLICE_F_NONBLOCK
was specified in
.I flags
or one of the file descriptors had been marked as nonblocking
.RB ( O_NONBLOCK ) ,
and the operation would block.
.TP
.B EBADF
One or both file descriptors are not valid,
or do not have proper read-write mode.
.TP
.B EINVAL
The target filesystem doesn't support splicing.
.TP
.B EINVAL
The target file is opened in append mode.
.\" The append-mode error is given since Linux 2.6.27; in earlier kernels,
.\" splice() in append mode was broken
.TP
.B EINVAL
Neither of the file descriptors refers to a pipe.
.TP
.B EINVAL
An offset was given for nonseekable device (e.g., a pipe).
.TP
.B EINVAL
.I fd_in
and
.I fd_out
refer to the same pipe.
.TP
.B ENOMEM
Out of memory.
.TP
.B ESPIPE
Either
.I off_in
or
.I off_out
was not NULL, but the corresponding file descriptor refers to a pipe.
.SH STANDARDS
Linux.
.SH HISTORY
Linux 2.6.17,
glibc 2.5.
.P
In Linux 2.6.30 and earlier,
exactly one of
.I fd_in
and
.I fd_out
was required to be a pipe.
Since Linux 2.6.31,
.\" commit 7c77f0b3f9208c339a4b40737bb2cb0f0319bb8d
both arguments may refer to pipes.
.SH NOTES
The three system calls
.BR splice (),
.BR vmsplice (2),
and
.BR tee (2),
provide user-space programs with full control over an arbitrary
kernel buffer, implemented within the kernel using the same type
of buffer that is used for a pipe.
In overview, these system calls perform the following tasks:
.TP
.BR splice ()
moves data from the buffer to an arbitrary file descriptor, or vice versa,
or from one buffer to another.
.TP
.BR tee (2)
"copies" the data from one buffer to another.
.TP
.BR vmsplice (2)
"copies" data from user space into the buffer.
.P
Though we talk of copying, actual copies are generally avoided.
The kernel does this by implementing a pipe buffer as a set
of reference-counted pointers to pages of kernel memory.
The kernel creates "copies" of pages in a buffer by creating new
pointers (for the output buffer) referring to the pages,
and increasing the reference counts for the pages:
only pointers are copied, not the pages of the buffer.
.\"
.\" Linus: Now, imagine using the above in a media server, for example.
.\" Let's say that a year or two has passed, so that the video drivers
.\" have been updated to be able to do the splice thing, and what can
.\" you do? You can:
.\"
.\" - splice from the (mpeg or whatever - let's just assume that the video
.\"   input is either digital or does the encoding on its own - like they
.\"   pretty much all do) video input into a pipe (remember: no copies - the
.\"   video input will just DMA directly into memory, and splice will just
.\"   set up the pages in the pipe buffer)
.\" - tee that pipe to split it up
.\" - splice one end to a file (ie "save the compressed stream to disk")
.\" - splice the other end to a real-time video decoder window for your
.\"   real-time viewing pleasure.
.\"
.\" Linus: Now, the advantage of splice()/tee() is that you can
.\" do zero-copy movement of data, and unlike sendfile() you can
.\" do it on _arbitrary_ data (and, as shown by "tee()", it's more
.\" than just sending the data to somebody else: you can duplicate
.\" the data and choose to forward it to two or more different
.\" users - for things like logging etc.).
.\"
.P
.B _FILE_OFFSET_BITS
should be defined to be 64 in code that uses non-null
.I off_in
or
.I off_out
or that takes the address of
.BR splice ,
if the code is intended to be portable
to traditional 32-bit x86 and ARM platforms where
.BR off_t 's
width defaults to 32 bits.
.SH EXAMPLES
See
.BR tee (2).
.SH SEE ALSO
.BR copy_file_range (2),
.BR sendfile (2),
.BR tee (2),
.BR vmsplice (2),
.BR pipe (7)