From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH] phonet: Check input from user before allocating Date: Tue, 03 Apr 2012 04:15:04 +0200 Message-ID: <1333419304.18626.5.camel@edumazet-glaptop> References: <201204022201.41808.remi@remlab.net> <20120402.174006.70116829696492189.davem@davemloft.net> <1333417997.18626.1.camel@edumazet-glaptop> <20120402.215911.1929019308299701014.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: remi@remlab.net, levinsasha928@gmail.com, remi.denis-courmont@nokia.com, davej@redhat.com, netdev@vger.kernel.org, linux-kernel@vger.kernel.org To: David Miller Return-path: In-Reply-To: <20120402.215911.1929019308299701014.davem@davemloft.net> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On Mon, 2012-04-02 at 21:59 -0400, David Miller wrote: > From: Eric Dumazet > Date: Tue, 03 Apr 2012 03:53:17 +0200 > > > Not that AF_UNIX does nothing in this respect, it can use order-XX pages > > for large datagrams. > > > > (I beleve I sent a patch some time ago to address this point) > > Yes, on the datagram side it's a problem. > > For stream AF_UNIX sockets the allocation is capped at SKB_MAX_ALLOC > which evaluates to an order 2 page. > > Overall, AF_UNIX ought to be easy to deal with since all of the > routines that copy data between userspace and SKBs can handle > segmented SKBs and thus most of the work is converting over to > sock_alloc_send_pskb() and setting data_len how we set the normal > length of sock_alloc_skb_skb() currently. > > Anyways, feel free to resubmit your patch. This was indeed a basic patch, but it probably can lower raw performance on some apps, (if memory frag is not an issue) so I need to bench it. Any idea of a representative benchmark in dgram af_unix ? http://patchwork.ozlabs.org/patch/114103/ I'll respin it with proper performance resuts.