Re: may be a buffer overflow problem

Hi,

On 2024-06-17 22:42:41 -0400, Tom Lane wrote:
> Andres Freund <andres@anarazel.de> writes:
> > On 2024-06-17 23:52:54 +0200, Daniel Gustafsson wrote:
> >> Since sqlca is, according to our docs, present in other database systems we
> >> should probably keep it a 5-char array for portability reasons.  Adding a
> >> padding character should be fine though.
> 
> > How about, additionally, adding __attribute__((nonstring))? Wrapped in an
> > attribute, of course.  That'll trigger warning for many unsafe uses, like
> > strlen().
> 
> What I was advocating for is that we make it *safe* for strlen, not
> that we double down on awkward, non-idiomatic, unsafe coding
> practices.

Given that apparently other platforms have it as a no-trailing-zero-byte
"string", I'm not sure that that is that clearly a win. Also, if they just
copy the field onto the stack or such, they'll have the same issue again.

And then there is this:

> Admittedly, I'm not sure how we could persuade compilers that
> a char[5] followed by a char field is a normal C string ...

I think the explicit backstop of a zero byte is a good idea, but I don't think
we'd just want to rely on it.

Greetings,

Andres Freund