Peter Maydell <peter.mayd...@linaro.org> wrote on 2014/07/11 18:00:25:
> On 11 July 2014 16:18, Joakim Tjernlund <joakim.tjernl...@transmode.se>
wrote:
> > Used by AF_PACKET sockets
> >
> > Signed-off-by: Joakim Tjernlund <joakim.tjernl...@transmode.se>
> > ---
> > linux-user/syscall.c | 8 ++++++++
> > linux-user/syscall_defs.h | 10 ++++++++++
> > 2 files changed, 18 insertions(+)
> >
> > diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> > index 1380f4e..a0e1ccc 100644
> > --- a/linux-user/syscall.c
> > +++ b/linux-user/syscall.c
> > @@ -1140,6 +1140,14 @@ static inline abi_long
target_to_host_sockaddr(struct sockaddr *addr,
> >
> > memcpy(addr, target_saddr, len);
> > addr->sa_family = sa_family;
> > + if (sa_family == AF_PACKET) {
> > + struct target_sockaddr_ll *target_lladdr, *lladdr;
> > + target_lladdr = (void *)target_saddr;
>
> Casting to (struct target_sockaddr_ll *) would be nicer
> than the void* cast.
>
> > + lladdr = (void *)addr;
>
> This looks fishy -- shouldn't lladdr be a struct sockaddr_ll, not
> a struct target_sockaddr? Alternatively, if (as we seem to)
> we rely on host and target layouts being the same, you might
> as well use struct sockaddr_ll in both cases and add a comment
> about our assumption.
>
> > +
> > + lladdr->sll_ifindex = tswap32(target_lladdr->sll_ifindex);
> > + lladdr->sll_hatype = tswap16(target_lladdr->sll_hatype);
>
> In fact we rely on memcpy() above to do most of the
> data copying, so you might as well just say
> /* data already copied, just swap the fields which might
> * be different endianness in host and guest
> */
> lladdr->sll_ifindex = tswap32(lladdr->sll_ifindex);
> lladdr->sll_hatype = tswap16(lladdr->sll_hatype);
>
> and drop the target_lladdr entirely.
Yes of course! I had messed with the code so much back and forth to
make it work so I completely lost track.
Jocke
