From mboxrd@z Thu Jan 1 00:00:00 1970 From: Pablo Neira Ayuso Subject: Re: [PATCH 0/9] netfilter: Compress hook function signatures. Date: Sat, 4 Apr 2015 13:49:28 +0200 Message-ID: <20150404114928.GA3829@salvia> References: <20150403.220734.1272465850706339981.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: netdev@vger.kernel.org, netfilter-devel@vger.kernel.org, hannes@stressinduktion.org, jiri@resnulli.us To: David Miller Return-path: Content-Disposition: inline In-Reply-To: <20150403.220734.1272465850706339981.davem@davemloft.net> Sender: netdev-owner@vger.kernel.org List-Id: netfilter-devel.vger.kernel.org On Fri, Apr 03, 2015 at 10:07:34PM -0400, David Miller wrote: > > Currently netfilter hooks have a function signature that is huge and > has many arguments. This propagates from the hook entry points down > into the individual hook implementations themselves. > > This means that if, for example, we want to change the type of one of > these arguments then we have to touch hundreds of locations. > > The main initial motivation behind this is that we'd like to change > the signature of "okfn" so that a socket pointer can be passed in (and > reference counted properly) for the sake of using the proper socket > context in the case of tunnels whilst not releasing the top level user > socket from skb->sk (and thus releasing it's socket memory quota > usage) in order to accomodate this. > > This also makes it clear who actually uses 'okfn', nf_queue(). It is > absolutely critical to make this obvious because any user of 'okfn' > down in these hook chains have the be strictly audited for > escapability. Specifically, escapability of references to objects > outside of the packet processing path. And that's exactly what > nf_queue() does via it's packet reinjection framework. > > In fact this points out a bug in Jiri's original attempt to push the > socket pointer down through netfilter's okfn. It didn't grab and drop > a reference to the socket in net/netfilter/nf_queue.c as needed. > > Furthermore, so many code paths are simplified, and should in fact be > more efficient because we aren't passing in arguments that often are > simply not used by the netfilter hook at all. > > Further simplifications are probably possible, but this series takes > care of the main cases. > > Unfortunately I couldn't convert ebt_do_table() because ebtables is > complete and utter crap and uses ebt_do_table() outside of the hook > call chains. But that should not be news to anyone. > > Signed-off-by: David S. Miller Nice series. If you route them through net-next, please add: Acked-by: Pablo Neira Ayuso or I can place them in nf-next, your call. Regarding Jiri's okfn() signature changes, please let me know on those patches, they will simplify a patchset I'll submit soon. And count on me to help on that change too, of course. Thanks a lot David!