Re: [PATCH v4 2/2] kallsyms: Don't leak pointers for unresolved symbols

[Sebastian Andrzej Siewior]

On 2026-10-19 14:05:55 [+0300], Andy Shevchenko wrote:
> On Mon, Oct 19, 2026 at 12:50:25PM +0200, Sebastian Andrzej Siewior wrote:
> > On 2026-10-19 13:34:35 [+0300], Andy Shevchenko wrote:
> > > > +int snprintf_default_pointer(char *buf, unsigned int len, const void *ptr)
> > > > +{
> > > > +	struct printf_spec spec = {
> > > > +		.base = 10,
> > > > +		.precision = -1,
> > > > +		.field_width = -1,
> > > > +	};
> > > 
> > > Isn't it the same as default_dec_spec?
> 
> > Except for .field_width, yes.
> 
> You can update it there. I think in all other cases (I briefly checked, but
> haven't run any smoke test with KUnit) it should be an equivalent.

Okay. So lets give it a try.

> > > > +	unsigned int w;
> > > > +	char *endp;
> > > > +
> > > > +	endp = default_pointer(buf, buf + len, ptr, spec);
> > > > +	w = min_t(unsigned int, len - 1, endp - buf);
> 
> This also won't properly work for len = 0, I assume? So one can't make
> something like kasprintf() on top of this.
> 
> > > May we avoid min_t()?
> > 
> > min() complains about signedness (len unsigned and ptr diff signed).
> > endp is >= buf so it will be unsigned.
> > An alternative would be manual comparison if the buffer size has been
> > exceeded or not and act upon it but this looks simpler.
> 
> It looks simpler, it looks wrong :-)

Could we then exclude len == 0 and keep the remaining bits as-is?

> > > > +	buf[w] = '\0';
> > > > +	return endp - buf;
> > > > +}

Sebastian