[Opensim-dev] Proposal to eliminate the name, description and invType fields from the assets db table

[Frisby, Adam]

Snip:
---
...Moreover, if we adopt generation of UUIDs using a hashing algorithm
(e.g. SHA1) at some stage, we wouldn't even need to embed the UUID in
the name...
---
Unsnip:

A warning here if I may.

*DO NOT DO THIS*. Hashing to produce a UUID for the asset, while
tempting in a space-saving kind of way, actually is not secure. Allow me
to explain further:

There are many known attacks on MD5 (and SHA1) which allow a duplicate
hash to be produced by tacking on additional data at the end of your
intended data, so - say someone wants to replace the default avatar with
a flying phallus, it would be possible to do, by creating a specifically
targeted asset with some dummy data on the very end to produce a
duplicate hash. When uploaded, it would override the original UUID of
that asset and hence could be bad.

If you *really* want to produce hashes in this way, use additional
length hashes which have yet to be easily collided with. If using SHA, I
recommend using at least 256 bit, ideally 512. This however does
increase the size of the key and hence slow down searches. It is also
recommended that it be easily possible to substitute the hashing
algorithm with a improved one later on - since these kinds of attacks
are ever evolving.

Regards,

Adam



-----Original Message-----
From: opensim-dev-bounces at lists.berlios.de
[mailto:opensim-dev-bounces at lists.berlios.de] On Behalf Of Justin
Clark-Casey
Sent: Saturday, 21 June 2008 6:35 PM
To: opensim-dev at lists.berlios.de
Subject: [Opensim-dev] Proposal to eliminate the name, description and
invType fields from the assets db table

This is mainly a question for core developers, though other viewpoints 
are welcomed.

In the OpenSim assets database table currently there are eight fields

+-------------+-------------+------+-----+---------+-------+
| Field       | Type        | Null | Key | Default | Extra |
+-------------+-------------+------+-----+---------+-------+
| name        | varchar(64) | NO   |     |         |       |
| description | varchar(64) | NO   |     |         |       |
| assetType   | tinyint(4)  | NO   |     |         |       |
| invType     | tinyint(4)  | NO   |     |         |       |
| local       | tinyint(1)  | NO   |     |         |       |
| temporary   | tinyint(1)  | NO   |     |         |       |
| data        | longblob    | NO   |     |         |       |
| id          | varchar(36) | NO   | PRI |         |       |
+-------------+-------------+------+-----+---------+-------+

Of these, name, description and invType are not really used by OpenSim 
(there are a couple of places in the code that use invType, but these 
could easily use other data sources).  Name and description are a little

pointless for assets - the real names and descriptions are actually 
always kept in the inventory items which reference these assets. 
InvType is also information which is always duplicated in items.

Eliminating these fields would have the following advantages

1.  Make it easier to save and load and integrate assets with locations 
outside OpenSim.  If you don't need to maintain name, description or 
invType, it should be possible to save assets with the uuid embedded in 
the filename and an extension which shows the asset type (e.g. .jp2 for 
JPEG2000 textures).  The assets wouldn't require persistence of the 
local and temporary attributes.

Two immediate use cases are that elimination of these asset fields would

  make the creation and manipulation of archives and inventory libraries

much simpler.  There may also be integration benefits.  For instance, 
asset data files could be served directly by a stock webserver without 
having to wrap them in XML to provide the extra metadata).  Moreover, if

we adopt generation of UUIDs using a hashing algorithm (e.g. SHA1) at 
some stage, we wouldn't even need to embed the UUID in the name.

2.  Eliminate the code which maintains these fields, simplifying the 
codebase and making it easier to read.  This is a minor point, since the

amount of code isn't large.

3.  Reduce data storage requirements.  This is also a minor point, since

these fields will always be dwarfed by the blobs (although this might 
come into play much further down the line if blobs are moved out of the
db).

The one disadvantage I can think of is

1.  It becomes a little hard to get a sense of what's in your asset 
database by just browsing the assets table.  For instance, you get a 
rough sense of how many objects are in prim or avatar inventories by 
counting the number of times that 'Primitive' appears as a name... 
However, statistical information could still be derived (if desired) by 
analysing the type field.

In summary, I would like to see if the name, description and invType 
asset table fields can be removed at some point in the near future. 
Does anybody have a good or necessary reason to retain them?

-- 
justincc
Justin Clark-Casey
http://justincc.wordpress.com
_______________________________________________
Opensim-dev mailing list
Opensim-dev at lists.berlios.de
https://lists.berlios.de/mailman/listinfo/opensim-dev