Re: [PATCH v1 1/1] seq_file: Replace strncpy()+nul by strscpy()

[Kees Cook]

On Mon, Jul 17, 2023 at 04:09:23PM -0700, Kees Cook wrote:
> 
> On Mon, 17 Jul 2023 12:33:32 +0300, Andy Shevchenko wrote:
> > Privided seq_show_option_n() macro breaks build with -Werror
> > and W=1, e.g.:
> > 
> > In function ‘strncpy’,
> >     inlined from ‘ocfs2_show_options’ at fs/ocfs2/super.c:1520:3:
> > include/linux/fortify-string.h:68:33: error: ‘__builtin_strncpy’ output may be truncated copying 4 bytes from a string of length 4 [-Werror=stringop-truncation]
> >    68 | #define __underlying_strncpy    __builtin_strncpy
> >       |                                 ^
> > include/linux/fortify-string.h:151:16: note: in expansion of macro ‘__underlying_strncpy’
> >   151 |         return __underlying_strncpy(p, q, size);
> >       |                ^~~~~~~~~~~~~~~~~~~~
> > cc1: all warnings being treated as errors
> > 
> > [...]
> 
> Applied, thanks!
> 
> [1/1] seq_file: Replace strncpy()+nul by strscpy()
>       https://git.kernel.org/kees/c/c30417b20f49

Gah, I dropped this from my tree since it was actually wrong[1]. This is an
ugly corner case with strscpy vs strncpy: the cast be32 from hfs/hfsplus[2]
looks unterminated to strscpy, so it would return -E2BIG, but really
FORTIFY noticed the over-read (strscpy is correctly checking the 5th
byte for NUL).

So... I think we need to fix seq_show_option_n() using memcpy+NUL, drop
the ocfs2 usage, and clarify that the seq_show_option_n() docs mean
"n means _exactly_ n bytes"...

-Kees

[1] https://lore.kernel.org/lkml/0000000000000a88cb0600ccef54@xxxxxxxxxx/
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/hfsplus/options.c?h=v6.4#n221

-- 
Kees Cook