On Thu, May 12, 2022 at 5:18 AM Bernd Walter <ti...@cicely7.cicely.de> wrote: > > On Wed, May 11, 2022 at 07:13:17PM -0400, Nathan Hartman wrote: > > On Wed, May 11, 2022 at 5:19 AM Fotis Panagiotopoulos > > <f.j.pa...@gmail.com> wrote: > > > > > > I was thinking the same thing. > > > > > > Which layer is responsible for this check? > > > IP, UDP, or ICMP? > > > > > > Logically it would seem to me that UDP should do the check and trigger > > sending the ICMP unreachable reply when appropriate, though it seems > > that it unconditionally calls icmp_reply(), which contains the logic > > to *avoid* sending it when it's the ANY or BROADCAST address. In other > > words, it is currently is coded in the ICMP layer. > > > > Looking at struct net_driver_s in include/nuttx/net/netdev.h, I am > > reminded that there is also IPv6, which a grep reveals has a similar > > handler in the IPv6 ICMP handler in icmpv6_reply() in > > net/icmpv6/icmpv6_reply.c. > > > > There is a function net_is_addr_mcast(), but there is no > > net_is_addr_bcast() currently. I think that is the function that needs > > to be coded, and then called from icmp_reply(). > > There is no broadcast in IPv6. > Completely replaced by multicast.
Sorry, I was not clear: I mentioned IPv6 only because I noticed how icmpv6_reply() is coded; icmpv6_reply() checks if the address is a multicast address by calling net_is_addr_mcast(). In contrast, the IPv4 version, icmp_reply(), checks against the ANY or BROADCAST addresses by a simple comparison. To fix the issue found by the OP, I proposed that we need a net_is_address_bcast(), which will be similar in purpose to net_is_address_mcast(), but for broadcast rather than multicast addresses. Then, icmp_reply() will call net_is_address_bcast() instead of the above-mentioned simple comparison that it is currently doing. The proposed new function net_is_address_bcast() will be smart enough to check for other broadcast addresses besides 255.255.255.255 by consulting the netmask. Cheers, Nathan
