[PATCH 04/11] sio: fix passing negative size_t to sio_add_to_iov

Vladimir Davydov vdavydov.dev at gmail.com
Wed Dec 5 11:48:07 MSK 2018


On Wed, Dec 05, 2018 at 12:29:20AM +0300, Vladislav Shpilevoy wrote:
> 
> 
> On 03/12/2018 16:50, Vladimir Davydov wrote:
> > On Fri, Nov 30, 2018 at 06:39:36PM +0300, Vladislav Shpilevoy wrote:
> > > sio_add_to_iov moves struct iov position on a
> > > specified offset, positive or negative. But its offset
> > > argument has size_t type, which is unsigned. Make it
> > > be ssize_t.
> > > 
> > > This worked before thanks to how negative numbers are
> > > stored. For example, consider
> > > 
> > > uint8_t value = 100;
> > > uint8_t offset = -5;
> > > 
> > > Value is stored as  0110 0100.
> > > Offset is stored as 1111 1011. (Yes, 1011, not 1010).
> > > 
> > > Sum of the values above is 0001 0101 1111 - first quad
> > > overflows and is truncated, so the result is
> > > 0101 1111 = 95 - correct.
> > > ---
> > >   src/sio.h | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/src/sio.h b/src/sio.h
> > > index ab0a243cd..ff383aa36 100644
> > > --- a/src/sio.h
> > > +++ b/src/sio.h
> > > @@ -84,7 +84,7 @@ sio_move_iov(struct iovec *iov, size_t nwr, size_t *iov_len)
> > >    * to adjust to a partial write.
> > >    */
> > >   static inline void
> > > -sio_add_to_iov(struct iovec *iov, size_t size)
> > > +sio_add_to_iov(struct iovec *iov, ssize_t size)
> > >   {
> > >   	iov->iov_len += size;
> > 
> > 'iov_len' has type size_t so 'size' will be converted to size_t before
> > the operation, in other words this patch has, in fact, no effect.
> 
> It fixes corrupted logic - you should not accept negative numbers in
> positive types. Even if they are the same internally.
> 
> > 
> > Anyway, it's OK to apply unary minus to an unsigned variable: no matter
> > how integer types are stored, whether the machine uses two's-complement
> > or not, it should work so that (-x + x) equals 0.
> 
> I do not argue with it. But then why do we have signed types? Lets use
> unsigned everywhere (<sarcasm>).
> 
> It does not matter does a compiler allow to turn a number into
> the complement or not. Logic of storing negative numbers in
> positive types is corrupted by definition.

But we do it all the time when adding a negative value to an unsigned
variable. Nobody cares since it conforms to the C standard.

> 
> > 
> > That being said, I don't think we need this patch.
> > 
> 
> As you wish. Removed in v2.

Thanks.



More information about the Tarantool-patches mailing list