25.1 Introduction

When using signal-driven I/O, the kernel notifies us with a signal when something happens on a descriptor. Historically, this has been called asynchronous I/O, but the signal-driven I/O that we will describe is not true asynchronous I/O. The latter is normally defined as the process performing the I/O operation (say a read or write), with the kernel returning immediately after the kernel initiates the I/O operation. The process continues executing while the I/O takes place. Some form of notification is then provided to the process when the operation is complete or encounters an error. We compared the various types of I/O that are normally available in Section 6.2 and showed the difference between signal-driven I/O and asynchronous I/O.

The nonblocking I/O we described in Chapter 16 is not asynchronous I/O either. With nonblocking I/O, the kernel does not return after initiating the I/O operation; the kernel returns immediately only if the operation cannot be completed without putting the process to sleep.

POSIX provides true asynchronous I/O with its aio_XXX functions. These functions let the process specify whether or not a signal is generated when the I/O completes, and which signal to generate.

Berkeley-derived implementations support signal-driven I/O for sockets and terminal devices using the SIGIO signal. SVR4 supports signal-driven I/O for STREAMS devices using the SIGPOLL signal, which is then equated to SIGIO.