From: Stefan Tomanek <redacted>
Date: Thu, 1 Aug 2013 02:24:07 +0200

Dies schrieb David Miller (davem@davemloft.net):

quoted

I just want to mention that the more quirky crap we put into the FIB
rules layer, the harder it will every be to make a scalable data
structure for FIB rule handling.

Right now it's basically a linear walk of rules, with processing at
each level.

And it still is: but instead of just having pre-conditions whether to
consult a table, the patch introduces post-conditions that can reject
a routing decision retrieved from it.

This doesn't change my argument at all.  The fact remains that the
more complex conditions we add to the fib rule lookup, the harder it
will be to optimize fib rule lookups into an O(1) or O(log n)
operation.