Thanks for so quick reply. Do you mean if (s->channel->afi == BGP_AF_IPV4)
maybe have bug, need change to if (s->channel->afi == BGP_AF_IPV4 &&
!s->channel-> ext_next_hop) ?
From: Ondrej Zajicek <santi...@crfreenet.org>
Sent: Wednesday, April 11, 2018 5:26 PM
To: Arvin Gan <a...@advaoptical.com>
Subject: Re: BGP attribute ext-next-hop
On Wed, Apr 11, 2018 at 07:55:28AM +0000, Arvin Gan wrote:
> Hi ,
> Thanks for your response. Could you help me for another question.
> If ext-next-hop is enabled and AFI is BGP_AF_IPV4, next- hop is a IPV4
> address, when update message created, the function bgp_create_mp_reach
> is called, what is the format of UPDATE message ? Firstly,
> bgp_encode_attrs function encode all attributes, is the attribute
> BA_NEXT_HOP encoded in this step ?
In traditional BGP (encoded by bgp_create_ip_reach() function), NEXT_HOP is
encoded like other attributes by bgp_encode_next_hop() function from attrs.c
file as part of bgp_encode_attrs().
In MP-BGP (encoded by bgp_create_mp_reach() function), NEXT_HOP attribute
should not be encoded independently as part of bgp_encode_attrs(), but is
encoded by bgp_encode_next_hop_ip() function (or its variant based on SAFI),
called through dispatch function directly from bgp_create_mp_reach()).
But now when i checked it, it seem like the condition in
bgp_encode_next_hop() does not take in to account ext-next-hop option.
So it is possible there is a bug here.
> If yes, as comment said, IPV4 NLRI is not used; If no,
> bgp_encode_next_hop assert next-hop is 16 bytes or 32 bytes, but IPV4
> next-hop is 4 bytes....
The assert in bgp_encode_next_hop_ip() function checks the length of the
internal representation of the next hop, which is always 16 bytes (type
ip_addr), even if IPv4 next hop is used (in such case it is encoded as
IPv4-mapped IPv6 address). Function ipa_to_ip4() gets 4B ip4_addr from 16B
Elen sila lumenn' omentielvo
Ondrej 'Santiago' Zajicek (email: santi...@crfreenet.org) OpenPGP encrypted
e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to
blame it on a computer is even more so."