RFC5722 and network packet duplication

[Dennis Ferguson <[e-mail]>]

IP should be tolerant of packet duplication in the network since this
is sometimes hard to avoid with certain link layer technologies and
can apparently also be caused in others by misconfiguration (I have
seen this happening without knowing the cause).  RFC5722 seems to
spend no text on the distinction between "overlapping fragments"
and "duplicate fragment packets" even though being able to differentiate
between the two is essential to keeping the fragmentation protocol robust
when the network is duplicating packets (there is a reason why the basic
reassembly procedures accept fragments carrying duplicate data...).

I was wondering if adding a definition of "overlapping fragments"
at the start of Section 2. to explicitly exclude duplicated packets
would make this better, maybe like:

   Two fragments of the same packet having different Fragment Offsets
   or different lengths of data after the fragment header are
   overlapping fragments if the offset ranges of the fragment data
   they carry overlap.   Duplicate fragment packets with the same
   Fragment Offset and length are not overlapping fragments and may
   be the result of packet duplication in the network rather than being
   intentionally constructed by the sender.

This unfortunately implies that reassembly implementations might need
to record the original packet boundaries, while the ones I've looked
at only track the data they've received and the holes they need to fill
to complete the packet.  I think, however, that the latter is sufficient
to avoid the attack 5722 imagines if the following rules are observed(?)

- if an incoming packet fragment only carries data that has already
  been received discard that packet fragment but continue the reassembly.

- if an incoming packet fragment includes both new data and data already
  received discard everything.

but this would be too big a change to make to the existing text.  I am
assuming RFC5722 did not intend to make the fragmentation protocol
intolerant of network packet duplication, but the fact that there is
clearly some implementation complexity involved with this that 5722
gives no attention to is making it hard to see exactly how this is
supposed to work.

Dennis Ferguson