Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
To:     linux-kernel@vger.kernel.org
From:   "Andrey Jr. Melnikov" <temnota.am@gmail.com>
Subject: Re: [PATCH] netfilter: nf_queue: Replace conntrack entry
Date:   Thu, 9 Aug 2018 17:09:37 +0300
Lines:  36
Message-ID: <vhut3f-tel.ln1@banana.localnet>
References: <20180503140745.26588-1-kristian.evensen@gmail.com>
User-Agent: tin/2.2.1-20140504 ("Tober an Righ") (UNIX) (Linux/4.3.3-bananian (armv7l))
Cc:     netfilter-devel@vger.kernel.org, netdev@vger.kernel.org
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

In gmane.comp.security.firewalls.netfilter.devel Kristian Evensen <kristian.evensen@gmail.com> wrote:
> SKBs are assigned a conntrack entry before being passed to any NFQUEUEs,
> and if no entry is found then a new one is created. This behavior causes
> problems for some traffic patterns. For example, if two UDP packets
> to/from the same host (using the same ports) arrive at the "same" time,
> both are assigned a new conntrack entry. After the first packet have
> traversed all chains, the conntrack entry will be inserted into the
> global table. The second packet will then be dropped during the
> insertion step, as an entry for the same flow already exists. One type
> of application that frequently generates this traffic pattern, is DNS
> resolvers.

> This commit introduces a new function that checks, and potentially
> replaces, the conntrack entry for any additional "new" SKBs mapping to
> an existing flow. While not a perfect solution, there are still
> situations where to-be-dropped SKBs can slip through, the situations is
> improved considerably. On the routers I have used for testing, packets
> belonging to the same UDP flow are let through (when generating the
> traffic pattern described above). Without the change in this commit, all
> packets except the first one was dropped.

> With the change in this commit, a user can implement "perfect" solutions
> in user-space. An application can for example keep track of seen UDP
> flows, and then only release packets belonging to one flow when the
> entry has been created. Without the change, and SKB is stuck with the
> original conntrack entry.

PING
Any progress on this patch?

> Signed-off-by: Kristian Evensen <kristian.evensen@gmail.com>
> ---
>  net/netfilter/nfnetlink_queue.c | 68 +++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 68 insertions(+)

[...]