Sasha,

On Thu, May 18, 2017 at 02:13:51PM +0000, Alexander Vainshtein wrote:
> I have several questions dealing with single-hop IP BFD sessions over IPv6. 
> So far I have failed to find clear answers to these questions in RFC 
> 5881<https://tools.ietf.org/html/rfc5881>.
> 
> 
> 1.       Is it possible to use link-local source and destination IPv6 
> addresses in encapsulation of single-hop IP BFD Control packets? For the 
> reference:
> 
> a.       Section 4 of RFC 5881 explicitly states that link-local addresses 
> SHOULD NOT be used in encapsulation of BFD Echo packets, but it does not say 
> anything about BFD Control packets

In the case of echo, there's no guarantee that the traffic will pass through
exactly one end system.  It's likely that you *could* do this using link
locals with appropriate discipline, but Echo packet contents are out of
scope for BFD.

For async single-hop sessions, you can do this.  The BFD implementation
obviously needs to take great care to send and receive on the same
interface.  Some socket implementations made such things a bit tricky in
years past.

> 
> b.      OSPFv3 for IPv6 always uses link-local IPv6 addresses as the Next Hop 
> addresses of the routes it computes (see RFC 
> 5340<https://tools.ietf.org/html/rfc5340>, Section 4.8.2). Therefore it looks 
> reasonable to me to use single-hop IPv6 BFD sessions with link-local 
> addresses at least for monitoring OSPFv3 adjacencies
> 
> 2.       Section 3 of RFC 5881 states that "there will be only a single BFD 
> session between  two systems over a given interface (logical or physical) for 
> a particular protocol". I would like to understand how this requirement can 
> be addressed in the following scenario:
> 
> a.       Let us assume that the answer to the question 1 above is positive.
> 
> b.      Let's further assume that:
> 
>                                                                i.      Router 
> A and Router B are connected across a single IPv6 hop (an IPv6 link)
> 
>                                                              ii.      A 
> single-hop IPv6 BFD session using link-local addresses of the corresponding 
> interfaces has been successfully established
> 
>                                                             iii.      
> Globally unique IPv6 addresses have been configured on the interfaces 
> terminating this link in Router A and Router B and have successfully passed 
> the DAD check, i.e., these addresses are assigned and preferred addresses in 
> the terminology of RFC 4862<https://tools.ietf.org/html/rfc4862>
> 
>                                                            iv.      The user 
> (or some application) now tries to set up a single-hop IPv6 BFD session bound 
> to the same interfaces but using globally unique IPv6 addresses assigned to 
> these interfaces

FWIW, the implementations I'm familiar with utilize the IP endpoints for the
session uniqueness requirements.  This means the global addresses would be
potentially on a different session than the link locals.

> c.       Should, under the assumptions above,  the implementation prevent 
> formation of an additional single-hop IPv6 BFD session between A and B 
> running across the same link but using assigned globally unique IPv6 
> addresses of the corresponding interfaces? If yes, how can this be achieved?

The difficulty in achieving this is the reason for my comment.

One could make a similar observation even in IPv4 when multiple addresses
are configured on the same interface.

> d.      Similar to above, but:
> 
>                                                                i.      The 
> interfaces connecting Router A and Router B have been assigned with multiple 
> globally unique IPv6 addresses
> 
>                                                              ii.      A 
> single-hop IPv6 BFD session using one pair of assigned IP addresses of these 
> interfaces has been successfully established
> 
>                                                             iii.      Should 
> the implementation prevent formation of an additional single-hop IPv6 BFD 
> session between A and B running on the same link but using a different pair 
> of assigned globally unique IP addresses? If yes, how can this be achieved?

Same as the IPv4.  I wouldn't expect this.

This arguably suggests the requirement in 5881 is excessively strict.

-- Jeff

Reply via email to