Re: Potentially false-positive -Wstringop-overflow= warning with gcc >= 11.1

[Segher Boessenkool <segher@kernel.crashing.org>]

On Fri, Jan 28, 2022 at 05:16:58PM +0100, Dumitru Ceara wrote:
> On 1/28/22 16:27, Segher Boessenkool wrote:
> > On Fri, Jan 28, 2022 at 04:01:36PM +0100, Dumitru Ceara via Gcc-help wrote:
> >>     void *l4data = p.l4_ofs != UINT16_MAX ? (char *) p.base_ + p.l4_ofs : NULL;
> >>     struct hdr2 *h2 = l4data;
> >>
> >>     memcpy(h2 + 1, &somedata[0], 6);
> > 
> > l4data can be 0, and everything false apart from there on.

(Wow, writing homonyms already, I must be tired)./

> In general, yes, l4data can be 0, if p.l4_ofs == UINT16_MAX, which can
> only happen if pkt_bar() changed p.base_.

Try it with the user code fixed though?  (Hint: the warning disappears).

> But the compiler can't know that for sure and the warning makes it sound
> like it's a sure thing:
> 
> "warning: ‘memcpy’ writing 6 bytes into a region of size 0 overflows the
> destination"

Yes.

These warnings are often questionable, especially on otherwise broken
code like this, but they are far from helpful then :-(

What perhaps happens here is that the compiler realises l4data is not 0
(because otherwise the program would have undefined behaviour).  But
your program explicitly makes that happen; perhaps the compiler should
have warned for that.  (It will do a runtime error if ever that path is
taken, instead, as things are).


Segher