| From: | Simon 'corecode' Schubert <corecode@xxxxxxxxxxxx> |
| Date: | Sat, 13 Sep 2003 22:06:40 +0200 |
Lately Matthew Dillon told: > My one issue is with the hash function... 32 bits isn't enough, I > would prefer a 64 bit hash (as a minimum). This is from my > experience with Diablo where even good 32 bit hash functions > produced collisions on news message-id's (and even 64 bit hash > functions in some cases). Actually I don't think this is much of a problem because each ELF object is also identified by its path component. Also, if I understood the paper right, the collision bound can easily be approximated by the number of hashed blocks. Let's take libc.so. As RTLD is hashing on DT_REL[A], we get about 13KB data. This is about 3.3e3 blocks (each 32 bit). With a collision bound at n*2e-28 we get a collision propability at 3.3e3*2e-28 = 6.6e-25 which is 1:1500000000000000000000000. But, of course, the hash function can be changed, or (what the authors suppose) used twice with different hash keys. cheers simon -- /"\ http://corecode.ath.cx/#donate \ / \ ASCII Ribbon Campaign / \ Against HTML Mail and News
Attachment:
pgp00012.pgp
Description: PGP signature