Re: [Ipsec] Minor comment/question on draft-ietf-ipsec-esp-v3-09.txt

>>>>> "Markku" == Markku Savela <msa@burp.tkv.asdf.org> writes:

 >> From: Stephen Kent <kent@bbn.com>

 >> Inside the ESP payload, after the optional IV, is the "next layer
 >> protocol header." It is referred to as "rest of payload data" in
 >> Figure 2. Typically, for transport mode this will be TCP or
 >> UDP. For tunnel mode, it is IP. This text is part of a discussion
 >> reminding folks that the IV, if present, must preserve the 4 or 8
 >> byte alignment requirements for the relevant IP version. We assume
 >> that the ESP payload, overall, started on an appropriate boundary,
 >> so the addition of the 8 bytes of SPI plus sequence number
 >> preserved this alignment. But the IV, if present, could break the
 >> alignment.

 Markku> I knew that, but I assumed the removal of the ESP headers
 Markku> from the packet required shifting of data within the packet
 Markku> anyway, so the aligment requirement seemed unnecessary.

No... you don't shift the data, you change the start of data pointer.
Then you reference header fields relative to that pointer.

If the alignment requirement isn't satisfied, then you can end up in
situations where the original alignment lets you do nice efficient
aligned memory loads, but after adjusting the pointer past the ESP
header you have to use unaligned load sequences.  Either that, or you
don't realize this and your box crashes on an unaligned load
fault... :-( 

	 paul

_______________________________________________
Ipsec mailing list
Ipsec@ietf.org
https://www1.ietf.org/mailman/listinfo/ipsec