Re: Re: uninitialized memory access in memmem()

[Rich Felker <dalias@libc.org>]

On Thu, Jun 19, 2014 at 01:43:50AM +0000, Clément Vasseur wrote:
> On 2014-06-19, Rich Felker <dalias@libc.org> wrote:
> > On Wed, Jun 18, 2014 at 06:20:33PM +0000, Clément Vasseur wrote:
> >> Hello,
> >> 
> >> I found a case where memmem() returns 0 where it should not:
> >> 
> >> $ cat test-memmem.c
> >> #define _GNU_SOURCE
> >> #include <string.h>
> >> #include <assert.h>
> >> 
> >> #define DATA 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10
> >> 
> >> int main(void)
> >> {
> >>     const unsigned char haystack[] = { DATA };
> >>     const unsigned char needle[] = { DATA };
> >>     assert(memmem(haystack, sizeof haystack, needle, sizeof needle));
> >> }
> >> 
> >> $ musl-gcc test-memmem.c && ./a.out
> >> Assertion failed: memmem(haystack, sizeof haystack, needle, sizeof needle) (test-memmem.c: main: 11)
> >> Aborted
> >> 
> >> Valgrind says a conditional jump or move depends on uninitalized value
> >> in twoway_memmem(). The code is quite complicated so I have not tried to
> >> track it down any further.
> >
> > Can you provide more details? musl version? gcc version? arch? I can't
> > reproduce this error in master with gcc 4.7.3/i386.
> 
> I use master (7c73cac) with gcc 4.6.1/x86_64.
> 
> I have another pattern which fails with gcc 4.8.3/arm. Looks like you
> might reproduce this one on your 32-bit arch:
> 
> #define DATA 0x50, 0x17, 0x8a, 0xf3, 0x55, 0x67, 0x58, 0xdf

Are you sure you're actually using musl master? The file that matters
is /lib/ld-musl-$ARCH.so.1. If it's a symlink to an old musl you have
lying around somewhere, then that's the version you're really using. I
was able to reproduce this with musl 1.0.0 and it's simply the bug
fixed in commit 476cd1d96560aaf7f210319597556e7fbcd60469.

Rich