I heard that FIFOs are named pipes. And they have exactly the same semantics. On the other hand, I think Unix domain socket is quite similar to pipe (although I've never made use of it). So I wonder if they all refer to the same implementation in Linux kernel. Any idea?

  • From the answer below I realized my question is sort of ambiguous and it's hard to be answered. It's likely no one could know so many details of implementational stuff in the kernel (even for kernel developers). If someone can confirm that Unix domain socket, pipe and FIFO they all buffer the data being sent in shared memory under Linux, my question is solved. Well... partly solved. – Justin May 15 '13 at 14:13
  • FIFO = named pipes != pipes. FIFOs can be bidirectional like a socket pair. Regular pipes are unidirectional. All have the file interface and file semantics. Why does implementation matter to you? – PSkocik Oct 21 '15 at 21:02
  • I know pipes are circular buffers, and that with the STREAMS system, these can have shared implementation, however Linux doesn't use STREAMS by default. I believe Linux hardcodes these IPC channels. I don't feel like checking, though. :D Why don't you? The code is publicly available. – PSkocik Oct 21 '15 at 21:04
  • If they all share the same implementation, their efficiency should be close to each other. And, to me, kernel code is too hard to understand. – Justin Oct 25 '15 at 12:59

UNIX domain sockets and FIFO may share some part of their implementation but they are conceptually very different. FIFO functions at a very low level. One process writes bytes into the pipe and another one reads from it. A UNIX domain socket has the same behaviour as a TCP/IP socket.

A socket is bidirectional and can be used by a lot of processes simultaneously. A process can accept many connections on the same socket and attend several clients simultaneously. The kernel delivers a new file descriptor each time connect(2) or accept(2) is called on the socket. The packets will always go to the right process.
On a FIFO, this would be impossible. For bidirectional comunication, you need two FIFOs, and you need a pair of FIFOs for each of your clients. There is no way of writing or reading in a selective way, because they are a much more primitive way to communicate.

Anonymous pipes and FIFOs are very similar. The difference is that anonymous pipes don't exist as files on the filesystem so no process can open(2) it. They are used by processes that share them by another method. If a process opens a FIFOs and then performs, for example, a fork(2), its child will inherit its file descriptors and, among them, the pipe.

The UNIX domain sockets, anonymous pipes and FIFOs are similar in the fact they use shared memory segments. The details of implementation may vary from one system to another but the idea is always the same: attach the same portion of memory in two distinct processes memory mapping to have them sharing data
(edit: that would one obvious way to implement it but that is not how it is actually done in Linux, which simply uses the kernel memory for the buffers, see answer by @tjb63 below).
The kernel then handles the system calls and abstracts the mechanism.

  • "UNIX domain sockets and FIFO may share some part of their implementation" ... the point is "some part of" ... I just realized my question is sort of ambiguous and it's hard to be answered. It's likely no one could know so many details of what parts they share in the kernel (even for kernel developers). Nevertheless... could anyone confirm that Unix domain socket, pipe and FIFO they all buffer the data being sent in shared memory under Linux? If it's confirmed, my question is solved. Well... partly solved. – Justin May 15 '13 at 14:09
  • Well, yes, there is a buffer managed by the kernel. For example, with FIFOs, yoou can kill the writer and the reader can still what was sent into the pipe before the death of the writer. With sockets, it's a bit more complicated because they function with a connected protocol. But if you send, say, an int to the socket, and then get out of scope so that the int is cleared from the sender stack, the receiver can still read it. So, there is clearly a buffer somewhere. – lgeorget May 15 '13 at 14:23
  • Re-reading the comment, not sure I'm clear here... Let me know if there still is something unclear. – lgeorget May 15 '13 at 14:29
  • Your comment is clear to me. – Justin May 15 '13 at 14:38

There's quite a good discussion of this here: http://www.slideshare.net/divyekapoor/linux-kernel-implementation-of-pipes-and-fifos

So far as I can see, both from the presentation slides, and the source @ http://lxr.free-electrons.com/source/fs/pipe.c - fifo's are implemented as a wrapper around pipe's, and pipe's themselves are implemented via the pipefs virtual filesystem..

@lgeorget - The pipes appear to use kernel memory for buffers between the readers and the writers - they don't use 'shared memory' as such, and copy memory between user and kernel address spaces (eg, pipe_read calls pipe_iov_copy_to_user, which calls __copy_to_user_inatomic (or copy_to_user). __copy_to_user_inatomic calls copy_user_generic, which is on of several ASM implementations.


A "FIFO" and a "named pipe" is the same thing - though it's quite different from how a shell handles a "pipe" (|) between two commands on the command-line.

A named pipe (FIFO) is a single "file" shared by two programs, where one writes to it and the other read from it... A socket on the other hand is a "connection" between two "files" - which may use a network and be on separate computers - where one program read/writes to one "file" and another program read/writes to the other... I don't think they're that similar... On the other hand both sockets and named pipes - as well as files, devices, symbolic links - all uses inodes, and they all implements some common features (like read and write).

  • 1
    Yes, Unix domain socket is kind of socket, so it's API is similar to other socket APIs such as TCP or UDP, etc. However, Unix domain socket can only be used as "local" IPC. And the way it transfers data is in first in first out manner, pretty much like FIFO & pipe. So I think it's possible Unix domain socket's API is just another encapsulation of identical implementation so we use it as if it were a socket. I think it's possible they all share the same internal in the kernel ... I want to confirm if it's true or not. – Justin May 15 '13 at 12:49

I don't think so Justin. If I'm not mistaken, and I quite possibly am, I think FIFOs use a file on disk, and Unix Domain sockets use kernel memory.

Also, as an addition to the poster above who mentioned that Unix domain sockets are bi-directional, that is only the case when using a SOCK_STREAM socket. SOCK_DGRAM Unix domain sockets are, in fact, uni-directional, and can only send() from the code that called connect(), to the code that called bind().

Of course, the code that called connect() must also call bind() to create it's own endpoint, but that has nothing to do with your question.

  • 3
    Welcome on StackExchange, and thank you for posting. A few remarks... 1) If you're "quite possibly" mistaken, you should double-check before answering ; this site is not a forum nor a chat. 2) Thank you for your precision on datagram oriented sockets 3 ) There is no need to post something that has "nothing to do" with the question. :) – lgeorget Jul 2 '13 at 20:51

My 2 cents...FIFO and UNIX socket are both bi-directional (similar) but socket have a star topology while a FIFO is just a queue (and hence cannot replace each other), yes their implementation may share code internally.


char * myfifo = "/tmp/myfifo";
mkfifo(myfifo, 0666);
fd = open(myfifo, O_RDWR);   //so that you can read/write to it
write(fd, buff1, sizeof(buff1));  
getchar();//wait till some one reds this and writes something else
int sz=read(fd, buff1, sizeof(buff1));  //read that something**
  • FIFO is bi-directional? – jhfrontz Aug 12 at 17:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.