The Picture You Already Know
Let’s get the basics on the table. You write code in 'user space,' a protected environment where your program can’t just go around touching hardware or messing with other processes. To do anything meaningful—like
read a file, send data over the network, or even get the current time—your application needs help from the boss: the operating system kernel. A system call, or 'syscall,' is the formal, secure process for asking the kernel to perform one of these privileged actions on your behalf. Your code packages up a request, transitions from user mode to kernel mode, the kernel does the work, and then it returns the result. You write file.read(), the kernel reads the disk. Simple enough. For many developers, the story ends there. But the real world is far messier.
The Interruption You Didn't Expect
Here's the hidden detail: a system call can be interrupted. Many syscalls are 'blocking,' meaning your program pauses and waits for the operation to complete. Think about reading data from a slow network connection or waiting for a user to type something into the terminal. Your program is essentially in a holding pattern, managed by the kernel.
But what happens if, during this wait, something else needs the kernel's attention? Specifically, what if a 'signal' is delivered to your process? A signal is a software interrupt, a notification sent by the OS to a process about an event. Maybe the user pressed Ctrl+C (SIGINT), or another process sent a signal to yours. When this happens, the kernel might have to stop what it's doing—your blocking system call—to handle the signal. The syscall doesn't fail, exactly. It just... stops. And this is where things get interesting.
Meet EINTR, Your New Frenemy
When a blocking system call is interrupted by a signal before it can complete, it returns an error. But it's a very special kind of error. On POSIX-compliant systems like Linux and macOS, the function will return -1, and the global error variable, errno, will be set to EINTR (short for 'Interrupted System Call').
This is the critical part most junior or self-taught developers miss: EINTR is not a fatal error. It’s not telling you 'the disk is full' or 'the network is down.' It’s the operating system’s polite way of saying, 'Sorry, I got distracted by a signal. Nothing is broken, but the thing you asked me to do isn't finished. You should probably try again.' If your code doesn't specifically check for EINTR and retry the call, it will likely treat it as a genuine failure, causing your program to crash or behave incorrectly.
How to Handle It Like a Pro
The correct way to handle this is to wrap your system calls in a loop that explicitly checks for EINTR. In C, the pattern is classic. Instead of just calling read(), you do something like this (in simplified form):ssize_t bytes_read;
while ((bytes_read = read(fd, buf, count)) == -1 && errno == EINTR) {
// The call was interrupted by a signal, so we just loop and try again.
continue;
}
Most modern, high-level languages like Python and Go often handle this for you automatically within their standard libraries, which is one reason you may have never encountered it directly. For example, Python's file I/O operations will typically retry on EINTR under the hood. However, if you are using lower-level libraries, interfacing with C code, or building highly resilient systems, knowing this pattern is non-negotiable. It's the difference between code that works 'most of the time' and code that works, period.
