Generating Unique Keys

[Paul Rubin]

trevp at trevp.net (Trevor Perrin) writes:
> I think that's a little iffy - SHA1 has a "length extension" property,
> where if you know h(m) for some message m that's a multiple of the
> block length (64 bytes in this case) you can compute h(m+x) (i.e. m
> with x appended).  So if one of your unique integers was a prefix of
> another, and your secret_prefix was the right size, you'd be in
> trouble.  It would be better to use hmac-sha1 with the secret_prefix
> as the key. 

We're talking about ordinary 32-bit integers, so it would take a
difficult combination of circumstances to make that attack work,
but yeah, using hmac would remove the issue.

> And it would be better to generate the secret_prefix as a good
> random number on system startup, not bake it into a configuration.

If you've got a source of good random numbers, you can just use them
directly as tokens and not need this hashing stuff.