On Thu, 25 Mar 2021 09:07:23 +0100, Kurt Van Dijck wrote: > On Thu, 25 Mar 2021 08:31:09 +0100, Oliver Hartkopp wrote: > > On 24.03.21 23:19, Richard Weinberger wrote: > > >Oliver, > > > > > >----- Ursprüngliche Mail ----- > > >>@@ -808,10 +810,13 @@ static int raw_recvmsg(struct socket *sock, struct msghdr > > >>*msg, size_t size, > > >> int noblock; > > >> > > >> noblock = flags & MSG_DONTWAIT; > > >> flags &= ~MSG_DONTWAIT; > > >> > > >>+ if (msg->msg_name && msg->msg_namelen < RAW_MIN_NAMELEN) > > >>+ return -EINVAL; > > >>+ This indeed fails. There exists a thin layer between the syscall and socket's recvmsg that puts a seperate sockaddr and length 0 :-( > > > > > >Like with Kurt's patch, my test fails here too because msg->msg_namelen is 0. > > >->msg_namelen is always 0 in the recvfrom() case. > > > > Agreed! > > > > I just adopted that part and did no more testing yesterday (needed some > > sleep). > > > > E.g. 'candump any' also fails at CAN frame reception time due to this wrong > > check ;-) > > I'd like to understand how this check is wrong. > msg->msg_namelen should contain the size of msg->msg_name, isn't it. > if that is less than RAW_MIN_NAMELEN, you can't store the address. > > candump sets msg->msg_namelen to sizeof(), so it is not 0. > > > > > Will send an update soon. > > looking forward > Kurt
