man2/splice.2

   1 .\" Hey Emacs! This file is -*- nroff -*- source.
   2 .\"
   3 .\" This manpage is Copyright (C) 2006 Jens Axboe
   4 .\" and Copyright (C) 2006 Michael Kerrisk <mtk.manpages@gmail.com>
   5 .\"
   6 .\" Permission is granted to make and distribute verbatim copies of this
   7 .\" manual provided the copyright notice and this permission notice are
   8 .\" preserved on all copies.
   9 .\"
  10 .\" Permission is granted to copy and distribute modified versions of this
  11 .\" manual under the conditions for verbatim copying, provided that the
  12 .\" entire resulting derived work is distributed under the terms of a
  13 .\" permission notice identical to this one.
  14 .\"
  15 .\" Since the Linux kernel and libraries are constantly changing, this
  16 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  17 .\" responsibility for errors or omissions, or for damages resulting from
  18 .\" the use of the information contained herein.  The author(s) may not
  19 .\" have taken the same level of care in the production of this manual,
  20 .\" which is licensed free of charge, as they might when working
  21 .\" professionally.
  22 .\"
  23 .\" Formatted or processed versions of this manual, if unaccompanied by
  24 .\" the source, must acknowledge the copyright and authors of this work.
  25 .\"
  26 .TH SPLICE 2 2009-02-20 "Linux" "Linux Programmer's Manual"
  27 .SH NAME
  28 splice \- splice data to/from a pipe
  29 .SH SYNOPSIS
  30 .nf
  31 .B #define _GNU_SOURCE
  32 .B #include <fcntl.h>
  33
  34 .BI "long splice(int " fd_in ", loff_t *" off_in ", int " fd_out ,
  35 .BI "            loff_t *" off_out ", size_t " len \
  36 ", unsigned int " flags );
  37 .fi
  38 .SH DESCRIPTION
  39 .BR splice ()
  40 moves data between two file descriptors
  41 without copying between kernel address space and user address space.
  42 It transfers up to
  43 .I len
  44 bytes of data from the file descriptor
  45 .I fd_in
  46 to the file descriptor
  47 .IR fd_out ,
  48 where one of the descriptors must refer to a pipe.
  49
  50 If
  51 .I fd_in
  52 refers to a pipe, then
  53 .I off_in
  54 must be NULL.
  55 If
  56 .I fd_in
  57 does not refer to a pipe and
  58 .I off_in
  59 is NULL, then bytes are read from
  60 .I fd_in
  61 starting from the current file offset,
  62 and the current file offset is adjusted appropriately.
  63 If
  64 .I fd_in
  65 does not refer to a pipe and
  66 .I off_in
  67 is not NULL, then
  68 .I off_in
  69 must point to a buffer which specifies the starting
  70 offset from which bytes will be read from
  71 .IR fd_in ;
  72 in this case, the current file offset of
  73 .I fd_in
  74 is not changed.
  75 Analogous statements apply for
  76 .I fd_out
  77 and
  78 .IR off_out .
  79
  80 The
  81 .I flags
  82 argument is a bit mask that is composed by ORing together
  83 zero or more of the following values:
  84 .TP 1.9i
  85 .B SPLICE_F_MOVE
  86 Attempt to move pages instead of copying.
  87 This is only a hint to the kernel:
  88 pages may still be copied if the kernel cannot move the
  89 pages from the pipe, or if
  90 the pipe buffers don't refer to full pages.
  91 The initial implementation of this flag was buggy:
  92 therefore starting in Linux 2.6.21 it is a no-op
  93 (but is still permitted in a
  94 .BR splice ()
  95 call);
  96 in the future, a correct implementation may be restored.
  97 .TP
  98 .B SPLICE_F_NONBLOCK
  99 Do not block on I/O.
 100 This makes the splice pipe operations non-blocking, but
 101 .BR splice ()
 102 may nevertheless block because the file descriptors that
 103 are spliced to/from may block (unless they have the
 104 .B O_NONBLOCK
 105 flag set).
 106 .TP
 107 .B SPLICE_F_MORE
 108 More data will be coming in a subsequent splice.
 109 This is a helpful hint when
 110 the
 111 .I fd_out
 112 refers to a socket (see also the description of
 113 .B MSG_MORE
 114 in
 115 .BR send (2),
 116 and the description of
 117 .B TCP_CORK
 118 in
 119 .BR tcp (7))
 120 .TP
 121 .B SPLICE_F_GIFT
 122 Unused for
 123 .BR splice ();
 124 see
 125 .BR vmsplice (2).
 126 .SH RETURN VALUE
 127 Upon successful completion,
 128 .BR splice ()
 129 returns the number of bytes
 130 spliced to or from the pipe.
 131 A return value of 0 means that there was no data to transfer,
 132 and it would not make sense to block, because there are no
 133 writers connected to the write end of the pipe referred to by
 134 .IR fd_in .
 135
 136 On error,
 137 .BR splice ()
 138 returns \-1 and
 139 .I errno
 140 is set to indicate the error.
 141 .SH ERRORS
 142 .TP
 143 .B EBADF
 144 One or both file descriptors are not valid,
 145 or do not have proper read-write mode.
 146 .TP
 147 .B EINVAL
 148 Target file system doesn't support splicing;
 149 target file is opened in append mode;
 150 .\" The append-mode error is given since 2.6.27; in earlier kernels,
 151 .\" splice() in append mode was broken
 152 neither of the descriptors refers to a pipe; or
 153 offset given for non-seekable device.
 154 .TP
 155 .B ENOMEM
 156 Out of memory.
 157 .TP
 158 .B ESPIPE
 159 Either
 160 .I off_in
 161 or
 162 .I off_out
 163 was not NULL, but the corresponding file descriptor refers to a pipe.
 164 .SH VERSIONS
 165 The
 166 .BR splice ()
 167 system call first appeared in Linux 2.6.17.
 168 .SH "CONFORMING TO"
 169 This system call is Linux-specific.
 170 .SH NOTES
 171 The three system calls
 172 .BR splice (),
 173 .BR vmsplice (2),
 174 and
 175 .BR tee (2),
 176 provide userspace programs with full control over an arbitrary
 177 kernel buffer, implemented within the kernel using the same type
 178 of buffer that is used for a pipe.
 179 In overview, these system calls perform the following tasks:
 180 .TP 1.2i
 181 .BR splice ()
 182 moves data from the buffer to an arbitrary file descriptor, or vice versa,
 183 or from one buffer to another.
 184 .TP
 185 .BR tee (2)
 186 "copies" the data from one buffer to another.
 187 .TP
 188 .BR vmsplice (2)
 189 "copies" data from user space into the buffer.
 190 .PP
 191 Though we talk of copying, actual copies are generally avoided.
 192 The kernel does this by implementing a pipe buffer as a set
 193 of reference-counted pointers to pages of kernel memory.
 194 The kernel creates "copies" of pages in a buffer by creating new
 195 pointers (for the output buffer) referring to the pages,
 196 and increasing the reference counts for the pages:
 197 only pointers are copied, not the pages of the buffer.
 198 .\"
 199 .\" Linus: Now, imagine using the above in a media server, for example.
 200 .\" Let's say that a year or two has passed, so that the video drivers
 201 .\" have been updated to be able to do the splice thing, and what can
 202 .\" you do? You can:
 203 .\"
 204 .\" - splice from the (mpeg or whatever - let's just assume that the video
 205 .\"   input is either digital or does the encoding on its own - like they
 206 .\"   pretty much all do) video input into a pipe (remember: no copies - the
 207 .\"   video input will just DMA directly into memory, and splice will just
 208 .\"   set up the pages in the pipe buffer)
 209 .\" - tee that pipe to split it up
 210 .\" - splice one end to a file (ie "save the compressed stream to disk")
 211 .\" - splice the other end to a real-time video decoder window for your
 212 .\"   real-time viewing pleasure.
 213 .\"
 214 .\" Linus: Now, the advantage of splice()/tee() is that you can
 215 .\" do zero-copy movement of data, and unlike sendfile() you can
 216 .\" do it on _arbitrary_ data (and, as shown by "tee()", it's more
 217 .\" than just sending the data to somebody else: you can duplicate
 218 .\" the data and choose to forward it to two or more different
 219 .\" users - for things like logging etc.).
 220 .\"
 221 .SH EXAMPLE
 222 See
 223 .BR tee (2).
 224 .SH SEE ALSO
 225 .BR sendfile (2),
 226 .BR tee (2),
 227 .BR vmsplice (2),
 228 .BR feature_test_macros (7)