On 12 July 2014 09:31, Joakim Tjernlund <joakim.tjernl...@transmode.se> wrote: > Peter Maydell <peter.mayd...@linaro.org> wrote on 2014/07/11 19:02:30: >> I thought about that, but there's a corner case: >> the kernel does the clamping of the optlen before the >> copy_from_user(), which means if you have: >> [interface name] [unreadable memory] >> and optlen is long enough that optval_addr + optlen >> reaches into the unreadable memory, then QEMU will return >> EFAULT (whereas the native kernel implementation will >> succeed) unless we do the clamping of the optlen ourselves. > > I can live with that IMHO very minor flaw that I dont think is > going to matter in practice for simplicity and speed of QEMU. > It is your call though, do we go for exact emulation or can we > cut some corners?
In this case I would prefer to get it right: * it's purely localised to this function * it's not all that hard to get right * we've already done the hard work of looking at the kernel and determining the correct behaviour SO_BINDTODEVICE is not going to be on any guest's speed-critical fastpath anyway... thanks -- PMM
