[rbridge] fragmentation for original ethernet frames

[Eric Gray]

Joe,

	One small correction...

	I think in the discussion below, I may not have been
very clear.  When the lower layer is asked for a buffer for
sending a PDU, the "maximum size" PDU it returns is based
on presumed knowledge (for instance, from configuration) of
the legitimate maximum PDU size at that layer - after taking
into account the overhead that will be inserted by that 
layer.

	The higher layer does not control the maximum size of
the buffer, only the amount of the buffer that it uses.

	Hence there is no implication that the lower layer may
need to deal with a larger payload than it "believes" itself
capable of.

--
Eric

> 	Also, because of other considerations, many "stacked" protocols
> don't so much "subtract" for headers added at lower layers as make a
> request for a PDU buffer (from the next lower layer) and get one of
> a certain maximum size.  Among other things, this approach allows a
> lower layer to account for whatever over-head it requires without
> having to share this information with the higher layer (this also
> allows the stack to avoid multiple memcpy operations as the content
> is handed off from layer to layer).

That presumes the lower layer is capable of a larger payload; as noted 
in the pointer I forwarded in my response to Vishwas, that's not always 
the case, and has already become an issue for Q-in-Q.

So the issue isn't new for TRILL, but it does affect TRILL.

Joe