As of June 2009, iptables and its relatives are still bound by the kernel interfaces of the ip_tables etc. modules, which only provide means to get and set entire tables. Old-fashioned scripts execute /sbin/iptables over and over — once for each policy they want to set, once for each flush operation they attempt, and once for each rule they are adding to the system — spend a much higher time finishing execution than would normally be required. Every iptables invocation will download the ruleset from the kernel, perform the modification — in case of -A this is adding just one rule — and upload it back into the kernel. Repeat that n times, and you get a cost factor of O(n^2) to perform n operations.

There is also a security risk, a window of opportunity where packets might pass halfway through while a ruleset is copied back and forth. Setting the policy to DROP before will not solve this problem, as this novice example shows:

С тех пор ничего не поменялось.