Work the Shell

Understanding Exit Codes

Dave Taylor

Issue #197, September 2010

In the real world, things don't always work out as you hoped. That's what exit codes are for—letting you know when things went wrong.

Last month, we looked at signals, the rudimentary mechanism that processes on a Linux box can use to communicate events and state changes. We talked about how each of the signals can be sent manually to a running process with the kill command, and how shell scripts then can catch and respond to specific signals (though not all of them—some cannot be caught because they're actually handled by the operating system itself).

Analogous to signals, exit codes turn out to be an easy way for processes to communicate state back to the calling parent process, in a way that most Linux users just ignore. Not anymore—this month, we're going to take a closer look.

mv Exit Codes

Let's start with a simple Linux command that everyone's probably already mastered: mv, which moves a file or directory from one spot in the filesystem to another (and/or renames it).

As you know, you can generate errors if the target is missing, destination is missing and so on. Here's quick example:

$ mv missing ~/missing2
mv: cannot move `missing' to `.../missing2': No such file or directory

You see an error; obviously, it didn't work. Ah, but behind the scenes, a numeric “return code” variable has been set in the shell too, something you can test and respond to within a shell script. Check out this sequence:

$ echo $?
0
$ mv missing ~/missing2
mv: cannot move `missing' to `.../missing2': No such file or directory
$ echo $?
1
$ echo test me
test me
$ echo $?
0

If no error occurs when executing a command, the exit code (which we reference in the shell with the shorthand $?) will have the value of 0: no error. Now, if I run a command that fails, the exit code will have a nonzero value. In the case of the failing mv command above, the error code will have the value of 1. And, if I now run yet another command, one which runs without error, the error code is reset to zero.

Now, let's take a peek at the mv man page, paying particular attention to the latter part of the doc. Close examination reveals: “The mv utility exits 0 on success, and >0 if an error occurs.”

That's not so interesting, really. The grep command has more interesting diagnostics, actually: “Normally, exit status is 0 if selected lines are found and 1 otherwise. But the exit status is 2 if an error occurred, unless the -q or --quiet or --silent option is used and a selected line is found.”

There is a set of system exit codes that are defined, although it's possible you'll never need the information. Here's a list of the codes and their meanings:

  • 1: general errors

  • 2: misuse of shell builtins (pretty rare)

  • 126: cannot invoke requested command

  • 127: command not found error

  • 128: invalid argument to “exit”

  • 128+n: fatal error signal “n” (for example, kill -9 = 137).

  • 130: script terminated by Ctrl-C

I'd never actually seen this list until I started digging into the issue of exit codes, so you can continue on your merry shell-scripting path safely without worrying about the data above.

Utilizing Exit Codes

The most common situation in which you analyze and respond to an exit code is in error handling in a script.

Here's a simple snippet where you want to create a directory. If it fails, you want to output an error message and quit:

#!/bin/sh
mkdir /usr
echo \$? = $?
if [ $? -ne 0 ] ; then
  echo "mkdir /usr failed: we have an exit code of $?"
  exit 1
fi
echo "made the requested directory. Why is '/' world writable?"
exit 0

It turns out, there's a nuance to working with the $? that I've already alluded to—one that makes output statements like the first “echo” quite problematic. You can see why in the output:

$ ./test.sh
mkdir: /usr: File exists
$? = 1
made the requested directory. Why is '/' world writable?

Can you see what happened? The exit code = 1 immediately after the mkdir, which makes sense as /usr already exists, but when we again test the exit code in the conditional, it's not a nonzero value!

Why? Because at that point, it indicates the exit code of the echo statement, not the mkdir command. Oops.

You can verify this simply by commenting out the first echo statement, in which case you now see this as the command output:

$ !.
./test.sh
mkdir: /usr: File exists
mkdir /usr failed: we have an exit code of 0

That makes more sense, doesn't it?

Because this can be tricky, a common thing I see in really bulletproof shell scripts with lots of error handling is something like this:

#!/bin/sh
mkdir /usr
error=$?
if [ $error -ne 0 ] ; then
  echo "mkdir /usr failed: we have an exit code of $error"
  exit 1
fi

This is one instance where a local variable to hold a system or global variable makes a lot of sense, and it also lets you do things like have an error message show up on-screen and be handed off to something like syslog() to ensure that the admin sees it at some point.

Of course, error handling doesn't always just need to print a message and exit. Another scenario might be the following:

alternates='
   http://www.example.com/test.pdf
   http://www.example2.com/test.pdf
   http://www.example3.com/test.pdf
   '
gotit=0
for file in $alternates
do
  wget $file
  if [ $? -ne 0 ]; then
    echo "Unable to get $file
  else
    gotit=1
    break
  fi
done
...

Here, we try to retrieve a file from one of multiple alternate locations and exit the loop only when we succeed (or when we've run out of possibilities).

Hiding Error Messages

Now that you know how to capture and analyze the exit codes from system commands, what if you want to have the error message be one from your script, not one from the command itself?

That's done with another new shorthand notation: >&, which redirects the stderr/error output stream. Here's how I use that to hide all error messages from the mkdir command being used in our sample scripts:


mkdir /usr >& /dev/null

You also can use &> or 2>&1 instead of >&.

If you don't test the results of the command, of course, you seriously can hose things up, but this makes the output more elegant for sure:

$ ./test.sh
mkdir /usr failed: we have an exit code of 0

Hmmm...I'm still getting that false 0. Oh! I haven't added the code to save the exit code value as “error”. One slight tweak later and:

$ ./test.sh
mkdir /usr failed: we have an exit code of 1

That's more like it!

I'm going to call this a wrap for this month. Next month, I'll demonstrate how the exit command lets you send exit codes back to the calling program from procedures and functions, just as if they were separate Linux commands rather than part of the same shell script.

Dave Taylor has been hacking shell scripts for a really long time, 30 years. He's the author of the popular Wicked Cool Shell Scripts, and he can be found on Twitter as @DaveTaylor and more generally at www.DaveTaylorOnline.com.