The Linux Programming Interface

908 Chapter 44

can be used for reading and writing on the FIFO. Doing this rather subverts the I/O

model for FIFOs, and SUSv3 explicitly notes that opening a FIFO with the O_RDWR

flag is unspecified; therefore, for portability reasons, this technique should be

avoided. In circumstances where we need to prevent blocking when opening a

FIFO, the open() O_NONBLOCK flag provides a standardized method for doing so (refer

to Section 44.9).

Avoiding the use of the O_RDWR flag when opening a FIFO can be desirable for a

another reason. After such an open(), the calling process will never see end-of-

file when reading from the resulting file descriptor, because there will always

be at least one descriptor open for writing to the FIFO—the same descriptor

from which the process is reading.

Using FIFOs and tee(1) to create a dual pipeline

One of the characteristics of shell pipelines is that they are linear; each process in

the pipeline reads data produced by its predecessor and sends data to its successor.

Using FIFOs, it is possible to create a fork in a pipeline, so that a duplicate copy of

the output of a process is sent to another process in addition to its successor in the

pipeline. In order to do this, we need to use the tee command, which writes two

copies of what it reads from its standard input: one to standard output and the

other to the file named in its command-line argument.

Making the file argument to tee a FIFO allows us to have two processes simulta-

neously reading the duplicate output produced by tee. We demonstrate this in the

following shell session, which creates a FIFO named myfifo, starts a background wc

command that opens the FIFO for reading (this will block until the FIFO is opened

for writing), and then executes a pipeline that sends the output of ls to tee, which

both passes the output further down the pipeline to sort and sends it to the myfifo

FIFO. (The –k5n option to sort causes the output of ls to be sorted in increasing

numerical order on the fifth space-delimited field.)

$ mkfifo myfifo

$ wc -l < myfifo &

$ ls -l | tee myfifo | sort -k5n

(Resulting output not shown)

Diagrammatically, the above commands create the situation shown in Figure 44-5.

The tee program is so named because of its shape. We can consider tee as function-

ing similarly to a pipe, but with an additional branch that sends duplicate output.

Diagrammatically, this has the shape of a capital letter T (see Figure 44-5). In addi-

tion to the purpose described here, tee is also useful for debugging pipelines and

for saving the results produced at some intervening point in a complex pipeline.

Figure 44-5: Using a FIFO and tee(1) to create a dual pipeline

ls

wc

tee sort

FIFO