OSDev.org

we should clearly define if we want to build a router or a host system. Even if linux kernel offers both, i don't think a hobby os should address routing packets. If your host(desktop/server/whatever) OS makes it to a point it start routing packets, it's no longer a hobby OS, imho.

btw, if a non-multicast router receives a request for a multicast join/prune IGMP message, that message is simply dropped (or at best, the emitter receives a ICMP error message). Same occurs for a router that receives a multicast packet: the packet is dropped or leads to a "no route to host" because the router has no table entry about that destination.

what do you exactly call "the fields of Multicast" ?? afaik, the only thing that distinguish a multicast packet from an 'ordinary' packet is that its destination address belongs to the multicast addresses range (must be class E or something)
http://www.networksorcery.com/enp/protocol/ip.htm

Pype.Clicker wrote: btw, if a non-multicast router receives a request for a multicast join/prune IGMP message, that message is simply dropped (or at best, the emitter receives a ICMP error message). Same occurs for a router that receives a multicast packet: the packet is dropped or leads to a "no route to host" because the router has no table entry about that destination.

Ok, didn't know you could be that rude to them. In that case, just be able to detect what addresses are not around and dump them.

what do you exactly call "the fields of Multicast" ?? afaik, the only thing that distinguish a multicast packet from an 'ordinary' packet is that its destination address belongs to the multicast addresses range (must be class E or something)
http://www.networksorcery.com/enp/protocol/ip.htm

class D to be precise, or in modern terminology, 224.0.0.0/4. And the fields I'm referring to are indeed source and dest. Sorry about the confusion.

great. i'm glad we finally got that stuff clear for everyone

Anyway, if you're going to implent TCP/IP, be sure you're going to implent IPv6 also.
It hasn't to be done immediatly, but it is advised to do it

DennisCGc wrote: Anyway, if you're going to implent TCP/IP, be sure you're going to implent IPv6 also.
It hasn't to be done immediatly, but it is advised to do it

No. I wouldn't advise that. For sure i hope we'll all use IPv6 some day, but IPv4 is still there for a couple of decades. Trying to support both from the start would only lead to code confusion and hard-to-test systems (unless you can come with a full IPv6 testbed at home: routers, servers, clients etc.)

Keep it for ToDoDreams. Really. Or pick up a kernel that exists and offer it IPv6 if that's what you wish.

I should mention ( :-[ ) that it's pretty difficult to implent, since I read the RFC.
I advise to do that, for those people that wants that their OS runs on companies, and that.
And I should mention too, that is not neccesary for now, but for in the future.
IPv4 runs for a couple of years, I think, but migration should be neccesary then ::)

just found my favourite book about TCP and networking avl. online.
http://www.cs.arizona.edu/llp/book/book.html

The TCP/IP Illustrated books by Richard Stevens are supposed to be good.