Re: [DNSOP] Clarification question: compression pointers always to names earlier in the packet?

[Shane Kerr <[e-mail]>]

Viktor,

On 24/10/2018 11.01, Viktor Dukhovni wrote:
> 
> My reading of RFC 1035 is that DNS name "compression"
> via "pointers" is restricted to name strictly earlier
> in the DNS message:
> 
>     4.1.4. Message compression
> 
>     In order to reduce the size of messages, the domain system utilizes a
>     compression scheme which eliminates the repetition of domain names in a
>     message.  In this scheme, an entire domain name or a list of labels at
>     the end of a domain name is replaced with a pointer to a prior occurance
>                                                              ---------------
>     of the same name.
> 
> And yet, here and there I see mention of having to take care to avoid "loops",
> but loops are impossible in a monotone strictly decreasing sequence.

I think the concern is something like this:

offset  octet
100     3
101     'f'
102     'o'
103     'o'
104     192
105     100


104 and 105 together point to offset 100, which will cause a loop even 
though they point to an earlier label.

> Is there a later RFC that relaxes the constraint and allows pointers to names
> later in the message?  I'm having a bit of trouble finding the later text...
> 
> Secondarily, can the pointer point to some odd-ball location earlier in the
> message that is not semantically a label in its original context, but just
> happens to carry data that decodes as the desired label?  Or, are pointers
> only valid to prior locations that are corresponding labels in their original
> context?

I have seen code which does this (there was an early version of the Knot 
authoritative server which used non-label offsets if it happened to have 
the correct data). Certainly it is easily possible that RDATA contains 
valid labels which could be used to by compression.

I think that for performance reasons most DNS software producing packets 
will maintain some data structure of prior labels & names in a packet 
and only reference those in compression. However, anything consuming DNS 
packets should probably be prepared for arbitrary pointers to previous 
locations in the packet - and also handle loops somehow (the usual 
options apply: either explicitly recording locations visited in 
traversal or just maintaining a counter and failing at some point).

Cheers,

--
Shane