Talking about strerror (and why you should be cautious when using it)

If you have ever used C/C++ before, you must know a very important global variable errno (though it can be a macro). Through this variable, we can get information about the library function call (whether it succeeds? If it fails, what's the type of error?). However, errno is just a number, and we want to interpret its meaning. Thus, we need some functions for us to "interpret" it. That's the reason why strerror exists.

What is strerror

Let's first take a look at the signature of the function.

char* strerror( int errnum );

The function takes in an integer errnum and "returns a pointer to the textual description of the system error code errnum"[1].

The description is very vague, but we do know

  • We don't need to free the memory of the char * returned from this function
  • We don't need to make sure errnum is in the valid range of errno

Let's read the rest of the description:

The returned string must not be modified by the program but may be overwritten by a subsequent call to the strerror function. strerror is not required to be thread-safe. Implementations may be returning different pointers to static read-only string literals or may be returning the same pointer over and over, pointing at a static buffer in which strerror places the string.

So, why the string returned may be overwritten by subsequent calls? To better understand it, let's imagine you need to implement a function that maps some integers to error messages. What would you do? I believe most of you, including me, will define some const char* strings as messages and returns one of them to the user. As far as I know, most implementations simply return a const char* to the user.

So, where does the possible overwriting come from? Let's look at the signature again char* strerror( int errnum );. It's the errnum. Remember, we could pass anything that is an int to this function. What would happen if we pass an integer, not in the range of the errno? (Clearly, your system doesn't have that many error types).

Let's see what would happen with the following programs:

#include <stdio.h>
#include <string.h>

int main(void){
    printf("%s\n", strerror(-1));
    return 0;
}

In my macOS Monterey 12.6, I get Unknown error: -1. You can see that it outputs the integer that we pass to the function.

If you are familiar with C, you may immediately realize how we could implement it: a static buffer inside this function with some predefined const char* message strings. If we have a valid errnum, we return one of the message strings; otherwise, we construct one inside our buffer and return the address of the buffer.

That's the root of evil.

Solutions before C11

Imagine if you want to implement a multi-threaded web server. You use system functions to create a socket and read/write data. Suppose some bad things happen, and two of your threads need to use strerror to acquire the error messages. Now, you face this situation:

  • You know most implementations return a const char* for valid errno
  • However, in theory, "strerror is not required to be thread-safe" (because of the static buffer). So if some unwise people, when developing this function, copy the error message to a static buffer and return it, you will have no idea what error messages you will get.

So, how could we mitigate this (before C11) if we want a portable version of strerror?

  1. Create a wrapper function and use a mutex. There are two possible implementations:
    1. Before releasing the mutex, copy the string into a dynamically allocated buffer. This method requires the user to free the returned result.
    2. Accept a user-provided buffer as a parameter and take in another parameter which is the size of the buffer. Copy the string into the buffer based on the given size.
  2. You could use sys_nerr and sys_errlist. The first argument is the length of the sys_errlist, and the second is an array that stores pointers to error messages. The detailed usage is not provided here because it is "deprecated and declared inconsistently" in some systems[2].
  3. Use strerror_r (which is another evil function)

The evilness of strerror_r

strerror_r is defined by POSIX as int strerror_r(int errnum, char *buf, size_t buflen);.

However, it has a different, non-standard prototype in glibc versus in POSIX: char *strerror_r(int errnum, char *buf, size_t buflen);
Even if you don't define _GNU_SOURCE or any other of the feature macros, you'll still get the non-POSIX definition on Linux.
Ok, you say, no big problem. I don't expect strerror_r to fail, so I should be able to get by with code like this:

strerror_r(errnum, buf, sizeof(buf)); // do NOT do this!
printf("foobar failed: %s\n", buf);

But, as the comment indicates, you must not do this. The GNU function doesn't usually modify the provided buffer-- it only modifies it sometimes. If you end up with the GNU version, you must use the return value of the function as the string to print[3].

Solution after C11

We finally see some great news since C11. We now have a improvded function in C standards: errno_t strerror_s( char *buf, rsize_t bufsz, errno_t errnum );

...The message is copied into user-provided storage buf. No more than bufsz-1 bytes are written, and the buffer is always null-terminated. If the message had to be truncated to fit the buffer and bufsz is greater than 3, then only bufsz-4 bytes are written, and the characters "..." are appended before the null terminator.

So, use this function if your project uses C11 or above.

What about C++?

Sadly, C++ doesn't have strerror_s or strerror_r. The only thing we have is strerror. So, create a wrapper and use a mutex if you care about "perfect portability".

Also, you should avoid using <system_error> (at least avoid using the error associated with std::error_code) for the same reason since its internal mechanism uses strerror to convert the system error code into the error message[4].


  1. cppreference.com: strerror, strerror_s, strerrorlen_s ↩︎

  2. Mac OS X manual page for strerror(3) ↩︎

  3. 2012-08-12: A portable strerror_r ↩︎

  4. libstdc++ implementation of system_error ↩︎