Re: Permissible states for extension frames #591

[Ilari Liusvaara <[e-mail]>]

On Mon, Aug 11, 2014 at 03:03:24PM -0700, Martin Thomson wrote:
> We don't really say this, but the implication is that an extension
> frame can appear anywhere, for any stream. Worst case, you can have
> frames appearing with any stream number at any time.
> 
> Are we OK with this?
> 
> Or... do we want to limit the sending of extension frames on streams
> somehow. Note that the most permissibly constrained frame type is
> PRIORITY, which can appear in any state other than "idle". A similar
> constraint would be relatively easy to enact.

Just as IMO:

I don't think "out-of-the-blue" extension frames have any use in idle
state (since state must be idle afterwards as well). And letting those
frames modify the sub-state in some manner would likely be a DOS attack
waiting to happen.

But negotated extensions are different matter, since those might define
new stream-opening frame types (or something even more radical).


-Ilari