Hello,
I'm hitting a strange issue with GCC while trying to wrap and mock
snprintf() for unit testing purposes.
I'm setting -fno-builtin and -fno-inline, but
__builtin___snprintf_chk() is getting called instead of snprintf().
The net effect is that my __wrap__snprintf() never gets called. Do you
have any advice on how to prevent object size checking builtins from
being used, without setting a lower optimization level? I expected
-fno-builtin to take care of it.
Note that out of my whole multi-distro test bed, I've only observed
this issue on Ubuntu 22.04, Ubuntu 20.04, and Fedora 36 Power 9. I
don't have direct access to the test machines, but I spun up an Ubuntu
22.04 reproducer. It uses gcc version 11.2.0-19ubuntu1.
A known working Fedora 36 x86_64 system uses gcc version 12.1.1
20220507. On that system, snprintf() (or more accurately
__wrap_snprintf()) gets called as expected, instead of the builtin.
There are also RHEL 7 machines in the testbed, so it works fine on
older versions as well.
Minimal reproducer:
```
# cat test.c
#include <stdio.h>
void func(const char *s)
{
char buf[16];
snprintf(buf, 16, "%s", s);
printf("%s\n", buf);
}
int main(void)
{
func("hello world");
}
# gcc -g -O2 -fno-builtin -fno-inline test.c -o test
# gdb -q ./test
Reading symbols from ./test...
(gdb) b 5
Breakpoint 1 at 0x10a0: file test.c, line 9.
(gdb) r
Starting program: /root/git/pacemaker/test
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Breakpoint 1, main () at test.c:9
9 {
(gdb) s
10 func("hello world");
(gdb)
func (s=s@entry=0x55555555600b "hello world") at test.c:3
3 {
(gdb)
5 snprintf(buf, 16, "%s", s);
(gdb)
0x00005555555551b9 in snprintf (__fmt=<optimized out>, __n=<optimized
out>, __s=<optimized out>) at
/usr/include/x86_64-linux-gnu/bits/stdio2.h:71
71 return __builtin___snprintf_chk (__s, __n, __USE_FORTIFY_LEVEL - 1,
```
Thank you.
--
Regards,
Reid Wahl (He/Him)
Senior Software Engineer, Red Hat
RHEL High Availability - Pacemaker
