Many users are only familiar with pipes as links between multiple flows, but they can do much more than that. Pipes can help you transfer data between computers. In this article, I will show you how to use pipes to redirect data streams in the shell.

Channels

Whenever a process starts under Linux, it is automatically assigned three channels. These channels have system assignments that let you address them, and each has a starting and end point. Channel 0 (STDIN) reads data, channel 1 (STDOUT) outputs data, and channel 2 (STDERR) outputs any error messages. Channel 2 typically points to the same device as channel 1 (Figure 1).

Figure 1: The shell reads input from the keyboard (STDIN, channel 0) and outputs the results on screen (STDOUT, channel 1). Error messages are displayed via STDERR (channel 2).

The shell itself, a Unix process, also uses these three channels. Each of them can be addressed via a file descriptor representing the respective channel number. On Linux, the channels used here physically reside in the /proc/PID/fd directory, where PID is equivalent to the process ID of the process being examined.

The Bash shell most commonly used on Linux also has channel 255. To make sure job control is retained when redirecting this channel, the shell sets it to STDERR at startup time.

Redirection

A redirection reads the channels of a process from a different source or outputs them to a different target. The most common use cases involve finding a string on the error channel and redirecting error messages to the /dev/null device.

The call in line 1 of Listing 1 attempts to display the nonexistent /dev/pseudo/ directory, which generates an error message on channel 2. The call from line 5 adds a redirect from channel 2 to /dev/null to the command. The error message now no longer appears on the screen, but the return value of the command remains unchanged.

01 # ls -ld /dev/pseudo
02 ls: cannot access /dev/pseudo: No such file or directory
03 # echo $?
04 2
05 # ls -ld /dev/pseudo 2>/dev/null
06 # echo $?
07 2

Pipes

A pipe is a special kind of file that acts as first in, first out (FIFO) memory in interprocess communication. With FIFO, one process writes to the pipe, while another reads from it. The reading process retrieves the characters in the same order as the writing process stored them. If, say, process 1 writes the values 1 Z 2 Y 3 X 4 W 5 V to a pipe, process 2 reads them from the pipe in the same order. Linux uses two types of pipes: anonymous and named pipes.

Anonymous pipes connect commands using the pipe symbol (|). These pipes are called anonymous because the user doesn't get to see them at runtime under normal circumstances. Anonymous pipes reside in the /proc/PID/fd/ directory like the standard channels. Calling a command sequence temporarily generates this kind of pipe.

Named pipes can be created in the filesystem using the mkfifo command. They remain active until deleted again using the rm command. When you use named pipes, you have to take care of the redirection work yourself, whereas the shell does this automatically for anonymous pipes.