Vulnerability Development mailing list archives

Re: Secure coding in C (was Re: Administrivia #4883)


From: marcs () ZNEP COM (Marc Slemko)
Date: Fri, 21 Jan 2000 01:12:16 -0700


On Thu, 20 Jan 2000, Tellier, Brock wrote:

In message <<A HREF="mailto:Pine.LNX.4.03.10001161207550.7428-100000 () brian citynet 
net">Pine.LNX.4.03.10001161207550.7428-100000 () brian citynet net</A>> Brian Masney writes:
: On some UNIX systems, snprintf does not guarentee that it will nul
: terminate the string. I know on some older versions of libc5 (sorry,
: don't have an exact version), if the buffer you was writing to got to the
: max size you passed it, it would stop there without adding the nul. So,
: you'll run into problems later on if you pass it to a string
: function (like strcpy())

snprintf is *DEFINED* to NUL terminate the string.  Systems >that don't
do this are broken.  That's why it is used as widely as it >is.

From the Solaris 7 snprintf man page:

The snprintf() function is identical to sprintf()
with the addition  of the argument n, which specifies
the size of the buffer referred to by s. The buffer is terminated  with  the null byte only if space is available.
--------------------------------^^^^^^^^^^^^^^^^^^^^^^^^^^

Meaning that one shouldn't copy more than bufsize - 1 bytes or risk a bof later on.

Well then it is broken.

Well, the man page is at least.

The code seems to do the right thing, at least on 2.6:

hostname:/tmp$ cat foo.c
#include <stdio.h>

int main() {
        char buff[11];
        strcpy(buff, "0123456789");
        snprintf(buff, 3, "abcdefg");
        printf("%s\n", buff);
        return(0);
}
hostname:/tmp$ gcc -Wall foo.c -o foo
hostname:/tmp$ ./foo
ab
hostname:/tmp$

But for portable code you need to include your own snprintf anyway if you
want to use it, so you may as well use it everywhere.  We did in Apache
and never looked back.  Even more usefully, in Apache it is expanded to
allow a psprintf that returns a string dynamically allocated using
Apache's pool mechanism, so in places where dynamic allocation is
appropriate, it isn't even a consideration.

(the "where it is appropriate" bit is, however, important, sine simply
switching everything to dynamically allocate memory has obvious DoS
potential in some situations)


Current thread: