Return-Path: From: ebiederm@xmission.com (Eric W. Biederman) To: Eric Dumazet Cc: Sven Joachim , Greg Kroah-Hartman , linux-kernel@vger.kernel.org, stable@vger.kernel.org, Ding Tianhong , Eric Dumazet , "David S. Miller" , Andy Lutomirski , Karel Srot References: <20130402221104.163133110@linuxfoundation.org> <20130402221116.307254752@linuxfoundation.org> <87vc833kpf.fsf@turtle.gmx.de> <87k3ojnosa.fsf@xmission.com> <1365034777.13853.46.camel@edumazet-glaptop> <1365035424.13853.48.camel@edumazet-glaptop> Date: Wed, 03 Apr 2013 18:01:11 -0700 In-Reply-To: <1365035424.13853.48.camel@edumazet-glaptop> (Eric Dumazet's message of "Wed, 03 Apr 2013 17:30:24 -0700") Message-ID: <878v4zjei0.fsf@xmission.com> MIME-Version: 1.0 Content-Type: text/plain Subject: Re: [ 105/124] af_unix: dont send SCM_CREDENTIAL when dest socket is NULL Sender: linux-kernel-owner@vger.kernel.org List-ID: Eric Dumazet writes: > On Wed, 2013-04-03 at 17:19 -0700, Eric Dumazet wrote: > >> Well, yes, this commit fixes a real bug : We were coalescing two >> messages into a single one, even if the senders were different. > > By the way, the 'LSB' test program can be found here : > > https://lsbbugs.linuxfoundation.org/attachment.cgi?id=2144 And we have two sends and one recvfrom, and no loops. So the the problem must be a failure to coalesce messages. There is a race between creating the sending and receiving sockets. At a first glance that race looks like we put the cred on the first message and not on the second message because we are connected by the time the second messages is sent. Which would definitely cause a failure to coallesce messages. Eric