James Ko James Ko - 2 months ago 8
C Question

I want to interpret stdin as a binary file. Why is freopen failing on Windows?

TL;DR: Why does

freopen(NULL, "rb", stdin)
always fail on Windows?

I'm trying to re-implement a
base64
encoder in C that takes input from stdin and outputs the encoded equivalent into stdout. I had a problem in my previous post where
fread
was signalling EOF prematurely. Here is what my main method looks like:

int main(void)
{
unsigned char buffer[BUFFER_SIZE];
unsigned char base64_buffer[BASE64_BUFFER];

while (1)
{
TRACE_PUTS("Reading in data from stdin...");
size_t read = fread(buffer, 1, sizeof(buffer), stdin); /* Read the data in using fread(3) */

/* Process the buffer */

TRACE_PRINTF("Amount read: %zu\n", read);
TRACE_PUTS("Beginning base64 encode of buffer");
size_t encoded = base64_encode(buffer, read, base64_buffer, sizeof(base64_buffer));

/* Write the data to stdout */
TRACE_PUTS("Writing data to standard output");
...

if (read < sizeof(buffer))
{
break; /* We reached EOF or had an error during the read */
}
}

if (ferror(stdin))
{
/* Handle errors */
fprintf(stderr, "%s\n", "There was a problem reading from the file.");
exit(1);
}

puts(""); /* Output a newline before finishing */

return 0;
}


Essentially, it reads in data from stdin using
fread
, encodes to base64, writes it to stdout, then checks if EOF has been reached at the end of the loop.

When I piped the contents of a binary file to this app's stdin, it would only read a fraction of the total bytes in the file. For example:

$ cat /bin/echo | my_base64_program >/dev/null # only view the trace output
TRACE: C:/Users/James/Code/c/base64/main.c:23: Reading in data from stdin...
TRACE: C:/Users/James/Code/c/base64/main.c:28: Amount read: 600
TRACE: C:/Users/James/Code/c/base64/main.c:29: Beginning base64 encode of buffer
TRACE: C:/Users/James/Code/c/base64/main.c:43: Writing data to standard output
TRACE: C:/Users/James/Code/c/base64/main.c:23: Reading in data from stdin...
TRACE: C:/Users/James/Code/c/base64/main.c:28: Amount read: 600
TRACE: C:/Users/James/Code/c/base64/main.c:29: Beginning base64 encode of buffer
TRACE: C:/Users/James/Code/c/base64/main.c:43: Writing data to standard output
TRACE: C:/Users/James/Code/c/base64/main.c:23: Reading in data from stdin...
TRACE: C:/Users/James/Code/c/base64/main.c:28: Amount read: 600
TRACE: C:/Users/James/Code/c/base64/main.c:29: Beginning base64 encode of buffer
TRACE: C:/Users/James/Code/c/base64/main.c:43: Writing data to standard output
TRACE: C:/Users/James/Code/c/base64/main.c:23: Reading in data from stdin...
TRACE: C:/Users/James/Code/c/base64/main.c:28: Amount read: 569
TRACE: C:/Users/James/Code/c/base64/main.c:29: Beginning base64 encode of buffer
TRACE: C:/Users/James/Code/c/base64/main.c:43: Writing data to standard output

$ cat /bin/echo | wc -c
28352


As you can see
/bin/echo
is 28352 bytes long, but only ~2400 of them are being processed. I believe the reason is because stdin is not being considered a binary file, so certain control characters (like Control-Z as mentioned in the linked post's answer) were prematurely signalling EOF.

I took a look at the base64 source code and it looks like they're using
xfreopen
(which is just a wrapper for
freopen
) to tell
fread
to interpret stdin as binary. So I went ahead and did that before the while-loop:

if (!freopen(NULL, "rb", stdin))
{
fprintf(stderr, "freopen failed. error: %s\n", strerror(errno));
exit(1);
}


However, now my app always exits at that point with:

$ cat /bin/echo | my_base64_program
freopen failed. error: Invalid argument


So why is
freopen
at that point failing, when it works for
base64
? I'm using MinGW-w64 with GCC on Windows if that's relevant.

Answer

Why freopen() might fail in general

The C standard says:

If filename is a null pointer, the freopen function attempts to change the mode of the stream to that specified by mode, as if the name of the file currently associated with the stream had been used. It is implementation-defined which changes of mode are permitted (if any), and under what circumstances.

Presumably, your implementation doesn't allow the changes you're trying to make. On Mac OS X, for example, the man page for freopen() adds:

The new mode must be compatible with the mode that the stream was originally opened with:

  • Streams originally opened with mode "r" can only be reopened with that same mode.
  • Streams originally opened with mode "a" can be reopened with the same mode, or mode "w".
  • Streams originally opened with mode ``w'' can be reopened with the same mode, or mode "a".
  • Streams originally opened with mode "r+", "w+", or "a+" can be reopened with any mode.

With that said, on Mac OS X (where b is a no-op anyway), you'd be OK.

Why freopen() fails on Windows specifically

However, you're on Windows. You need to learn how to find and read the documentation. I use a Google search with the term 'site:msdn.microsoft.com freopen' for whatever function I'm looking for. That specific search yields the manual for freopen() where it says:

If path, mode, or stream is a null pointer, or if filename is an empty string, these functions invoke the invalid parameter handler, as described in Parameter Validation. If execution is allowed to continue, these functions set errno to EINVAL and return NULL.

That's the documented behaviour: it is also the behaviour you are seeing. The manual for your system is helpful. It basically says "thou shalt not".

How to fix the input mode of standard input on Windows

I note that in my answer to your previous question, I pointed to _setmode():

However, it is more likely that you need _setmode():

_setmode(_fileno(stdin), O_BINARY);

This is the advice that is given in the answers to the question that deamentiaemundi pointed to.

I note in passing that the Microsoft manual page for setmode() says:

This POSIX function is deprecated. Use the ISO C++ conformant _setmode instead.

This is a curious comment because POSIX does not standardize a function setmode() in the first place.

You can find Microsoft's documentation for fileno(). It too has the spiel about POSIX (but this time it is accurate; POSIX does specify fileno()) and refers you to _fileno().

Comments