Re: ip_conntrack_hash() problem

[Martin Wilck <Martin.Wilck@Fujitsu-Siemens.com>]

Just to make my previous statement clearer:

I think there's nothing wrong with a power-of-2 hashsize, as long as the
hash function contains no implicit or explicit multipliers that are also
powers of two. In general, multipliers should not have a greatest common
divisor (GCD) larger than 1 with the hash size. Unfortunately, in the
current implementation, ntohl() creates an implicit multiplier of 2^16
for the port numbers (on little-endian machines).

Martin

PS: For the sake of that, the patch also changed the multiplier for the
source port from 2 to 7, assuming that it's relatively unlikely to have
a hash size that is a multiple of 7, and knowing that  multiplying by 7
is cheap. Instead of 7, 31 or 127 also seem good candidates that are
even more unlikely to be divisors of the hash size.

I recommend to printk() a warning if the hash size turns out to have a
GCD >1 with multiple of any multiplier in the hash function.

-- 
Martin Wilck                Phone: +49 5251 8 15113
Fujitsu Siemens Computers   Fax:   +49 5251 8 20409
Heinz-Nixdorf-Ring 1	    mailto:Martin.Wilck@Fujitsu-Siemens.com
D-33106 Paderborn           http://www.fujitsu-siemens.com/primergy