Re: [PATCH v3 0/4] Make sscanf() stricter

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Mon, Jun 12, 2023 at 05:23:18PM -0400, Demi Marie Obenour wrote:
> On Tue, Jun 13, 2023 at 12:00:44AM +0300, Andy Shevchenko wrote:
> > On Mon, Jun 12, 2023 at 04:25:01PM -0400, Demi Marie Obenour wrote:
> > > On Mon, Jun 12, 2023 at 02:59:38PM +0300, Alexey Dobriyan wrote:
> > > > > +	bool _placeholder;
> > > > > +	return simple_strntoull(cp, INT_MAX, endp, base, &_placeholder);
> > > > 
> > > > This can be done without introducing dummy variables:
> > > > 
> > > > 	void f(bool *b)
> > > > 	{
> > > > 	}
> > > > 
> > > > 	f((bool[1]){});
> > > 
> > > This is more consise, but (at least to me) significantly less readable.
> > > 
> > > > > > lib/vsprintf.c:3727:26: error: unknown conversion type character ‘!’ in format [-Werror=format=]
> > > > > So NAK.
> > > > 
> > > > Yeah, ! should go after format specifier like it does for %p.
> > > 
> > > I hadn't considered that.  Is the typical approach in Linux to use e.g.
> > > %d%[!] if one wants a literal '!'?
> > 
> > It might be that the cleanest way we have is to create %p-like extensions to
> > sscanf(). %p takes alnum as parameter and that is usually works since it makes
> > a little sense to attach alnum suffix to the pointer.
> > 
> > (I don't like to have %dX, where X is alnum as we expanding our hack to
> >  something which people don't expect to be altered even in the kernelm, you may
> >  refer to the discussion about %de for printing errors)
> 
> Personally I’m not too worried about compatibility with userspace
> sscanf(), except to the extent that -Werror=format can keep working.
> Userspace sscanf() is almost useless: it has undefined behavior on
> integer overflow and swallows spaces that should usually be rejected.
> I typically either use strto*l() or (as I am currently doing for Xen’s
> toolstack) just write my own parsing functions from scratch.

`man sscanf` tells about %p, and currently we have no use (if I'm not mistaken)
for %pj in printf(), so that can be used for %pj in sscanf() to avoid ambiguity
with possible extensions to actually parse our %p extension-like strings.

Not sure if others support the idea.

-- 
With Best Regards,
Andy Shevchenko