Re: Undefined Behavior in rw_verify_area() (was: sendfile(2) erroneously yields EINVAL on too large counts)

[Alejandro Colomar <alx@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 1185 bytes --]

Hi Matthew!

On Mon, Mar 04, 2024 at 03:22:24PM +0000, Matthew Wilcox wrote:
> On Mon, Mar 04, 2024 at 04:09:26PM +0100, Alejandro Colomar wrote:
> > Depending on the width of those types, the sum may be performed as
> > 'loff_t' if `sizeof(loff_t) > sizeof(size_t)`, or as 'size_t' if
> > `sizeof(loff_t) <= sizeof(size_t)`.  Since 'loff_t' is a 64-bit type,
> > but 'size_t' can be either 32-bit or 64-bit, the former is possible.
> > 
> > In those platforms in which loff_t is wider, the addends are promoted to
> > 'loff_t' before the sum.  And a sum of positive signed values can never
> > be negative.  If the sum overflows (and the program above triggers
> > such an overflow), the behavior is undefined.
> 
> Linux is compiled with -fwrapv so it is defined.

Hmmm; thanks!  Still, I'm guessing that's used as a caution to avoid
opening Hell's doors, rather than a declaration that the kernel doesn't
care about signed-integer overflow bugs.  Otherwise, all the macros in
<linux/kernel/overflow.h> wouldn't make much sense, right?

Have a lovely day!
Alex

-- 
<https://www.alejandro-colomar.es/>
Looking for a remote C programming job at the moment.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]