Xtonlsab bug

[Joakim Tjernlund]

>
> Ondrej Zajicek <santiago at crfreenet.org> wrote on 2010/04/29 23:15:22:
> >
> > On Thu, Apr 29, 2010 at 11:03:32PM +0200, Joakim Tjernlund wrote:
> > >
> > > Ondrej, this looks buggy:
> > >
> > > +static inline void htonlsab(void *h, void *n, u16 len) { memcpy(n, h, len); };
> > > +static inline void ntohlsab(void *n, void *h, u16 len) { memcpy(h, n, len); };
> > >
> > > memcpy is not defined to handle overlapping memory. Best to add:
> > > if (n != h)
> > >   memcpy(...)
> >
> > Yes, but all usages of htonlsab()/ntohlsab() are non-overlapping.
> > (overlapping/on place usage was replaced with htonlsab1()/ntohlsab1()).
>
> I see, but it would be safer to have that check in case you
> used them in the wrong way?
>
> >
> > > Might as well do that for these too:
> > > +static inline void htonlsah(struct ospf_lsa_header *h, struct
> > ospf_lsa_header *n) { *n = *h; };
> > > +static inline void ntohlsah(struct ospf_lsa_header *n, struct
> > ospf_lsa_header *h) { *h = *n; };
> >
> > *n = *h should be OK even if n == h, AFAIK.
>
> Hopyfully, but isn't gcc allowed to use memcpy to do that
> assignment?
> If you add that if (h!=n) test, doesn't the need for Xlasb1() go away as well?

..and use memmove() instead. That way h == n becomes a NOP and
h!=n can be overlapping as memmove can do that.