Re: rfc4941bis: Invalid addresses used by ongoing sessions

[Fernando Gont <[e-mail]>]

On 10/2/20 17:25, Mark Smith wrote:
> 
> 
> On Tue, 11 Feb 2020, 03:12 Fernando Gont, <[e-mail] 
> <mailto:[e-mail]>> wrote:
> 
>     Folks,
> 
>     As currently specified, temporary addresses are removed when they become
>     invalid (i.e., the Valid Lifetime expires).
> 
>     Section 6 ("6.  Future Work") of the draft
>     (https://tools.ietf.org/id/draft-ietf-6man-rfc4941bis-06.txt) still
>     keeps the following text from RFC4941.
> 
>     6.  Future Work
> 
>         An implementation might want to keep track of which addresses are
>         being used by upper layers so as to be able to remove a deprecated
>         temporary address from internal data structures once no upper layer
>         protocols are using it (but not before).  This is in contrast to
>         current approaches where addresses are removed from an interface
>     when
>         they become invalid [RFC4862], independent of whether or not upper
>         layer protocols are still using them.  For TCP connections, such
>         information is available in control blocks.  For UDP-based
>         applications, it may be the case that only the applications have
>         knowledge about what addresses are actually in use. 
>     Consequently, an
>         implementation generally will need to use heuristics in deciding
>     when
>         an address is no longer in use.
> 
> 
>     I wonder if this text should be:
> 
>     1) moved more into the body of the document and made a "MAY" (which for
>     TCP is very straightforward),
> 
>     2) Be left "as is", or,
> 
>     3) Removed from the document
> 
> 
> I generally like the idea, however I think it should be removed at this 
> stage because of the ambiguity problem.
> 
> Just because an address has been used by TCP doesn't mean it isn't 
> currently or can't in the near future also be used by UDP (or DCCP or 
> SCTP). Current source address selection doesn't take into account the 
> transport layer protocol.

Note: this is simply an "heuristic"/hack to prevent long-lived 
connections from being torn down. i.e., by default,, they would be torn 
down when the address becomes invalid. This "hack" would avoid such 
scenarios.



> If the temporary address mechanism eventually evolved to the point where 
> there were "TCP only" and "UDP only" (and DCCP, SCTP) pools of temporary 
> addresses, then this would be possible to reliably implement.

If you don't implement this (which is the status quo), then when the 
address becomes invalid, any long-live TCP sessions using this address 
will be torn down. With this hack, that wouldn't happen. Connectionless 
protocols would remain unaffected.  -- just a clarification, so that 
we're on the same page.

Thanks!

Cheers,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492