0

I'm writing a PMTUD app for both IPv4 and v6. I am doing this on Ubuntu 12.04, but I would like to make it as OS-independent as possible, and that's where I stumbled upon a problem.

IPv6 packets get fragmented by the sender by default, and I do not know how to turn this behaviour off. I found some socket options like IPV6_MTU_DISCOVER and IPV6_DONTFRAG, but I found these under linux/in6.h, which does not help as I'm using the netinet header family and neither of those is under netinet/in.h - although IPV6_MTU_DISCOVER should be there according to this. Am I missing something?

EDIT: Let me clarify a bit then. I have a socket(AF_INET6, SOCK_RAW, IPPROTO_ICMPV6) through which I wish to send an ICMPv6 packet of such size that I will receive a reply telling me it's too big, and from that reply I will get the path MTU. However, to truly get the MTU along the whole path I also have to factor in the outgoing device's MTU.

I am using miredo to tunnel IPv6, which has a set MTU of minimal size, e.g. 1280. Sending a packet bigger that 1280 will result in fragmentation of said packet (this behaviour I observed in Wireshark), but I need the socket to REFUSE to send the packet and inform me about it rather than fragment it.

user742925
  • 275
  • 1
  • 2
  • 9
  • Could you explain a bit more what you are trying to do? When you say "IPv6 packets get fragmented by the sender by default", what are the packet sizes? What are the link MTUs? What sockets options are you using? – Nikolai Fetissov Nov 14 '12 at 13:44
  • What type of socket are you using to send your data? It sounds like you're using the IPv6 Raw socket to get the MTU and then using SOCK_STREAM for transmitting your actual data. This won't work, as SOCK_STREAM sockets will perform fragmentation for you. – Jeff Nov 16 '12 at 00:53

2 Answers2

3

You do not need to do this yourself. MTU discovery is supposed to happen automatically. As a side effect of this, all devices along the path MUST allow ICMP V6 packets to pass.

jm7
  • 31
  • 1
-2

IPv6 packets get fragmented by the sender by default

No. TCP packets get fragmented by the sender and intermediate routers by default.

, and I do not know how to turn this behaviour off.

You cannot turn it off. You can certainly try, but the only result will be non-delivery. If a router needs to fragment a packet and you don't permit it, it will drop it instead. However the sending host also needs to fragment, to fit inside the path MTU, and you cannot stop that. If you write the receiver correctly, i.e. in the expectation that it is reading a byte stream rather than discrete messages, it should make no difference to you whether the packets got fragmented in transit or not.

user207421
  • 305,947
  • 44
  • 307
  • 483
  • Updated the question, maybe I explained it badly. Non-delivery is the desired outcome - I NEED the sending host to drop the packet and tell me it did so instead of fragmenting to fit the MTU. That's the point of PMTUD, if I'm not mistaken. – user742925 Nov 15 '12 at 21:11
  • 1
    No. Routers only act on layer 3, so on legacy IPv4 or on modern IPv6, never on TCP. It's perfectly possible (and for a debugging app perfectly sensible) to do this yourself by specifying don't fragment for IPv4. IPv6 is never fragmented by intermediate routers anyway. – Kristof Provost Nov 22 '13 at 13:29