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

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

On Tue, Jun 20, 2023 at 04:57:55PM +0200, Petr Mladek wrote:
> On Tue 2023-06-20 16:52:42, Andy Shevchenko wrote:
> > On Tue, Jun 20, 2023 at 03:34:09PM +0200, Petr Mladek wrote:
> > > On Thu 2023-06-15 14:23:59, Andy Shevchenko wrote:
> > > > On Thu, Jun 15, 2023 at 08:06:46AM +0000, David Laight wrote:

...

> > >   + %pj or another %p modifiers would be hard to understand either.
> > > 
> > >     Yes, we have %pe but I think that only few people really use it.
> > >     And it is kind of self-explanatory because it is typically
> > >     used together with ERR_PTR() and with variables called
> > >     "err" or "ret".
> > 
> > j, besides the luck of no (yet) use in the kernel's printf(), is
> > described for printf(3)
> > 
> >   j   A  following integer conversion corresponds to an intmax_t or uintmax_t
> >       argument, or a following n conversion corresponds to a pointer to an
> >       intmax_t argument.
> 
> I see, I have missed this coincidence. And we would really need to use %pj.
> %jd requires intmax_t variable. Otherwise, the compiler produces:
> 
> kernel/lib/test.c:10:17: error: format ‘%jd’ expects argument of type ‘intmax_t *’, but argument 3 has type ‘int *’ [-Werror=format=]
>   sscanf(str, "%jd hello.", &tmp);
> 
> Hmm, %pj might even make some sense for sscanf() which requires pointers anyway.
> But still, we would lose the compiler check of the size of the passed
> buffer.
> 
> This is actually my concern with many other %p modifiers. The compiler
> is not able to check that we pass the right pointer. I know that this
> might happen even with wrong buffer passed to %s or so. But number
> types is another category.

Yeah, it was a discussion IIRC for the compiler plugin to support %p
extensions, but I have no idea where it's now.

> > So, I think among all proposals, this one is the best (while all of them may
> > sound not good).
> 
> I still prefer the custom handler when it is not too complex.
> 
> Or if there are many users, we could create sscanf_strict() or so.

-- 
With Best Regards,
Andy Shevchenko