From owner-freebsd-net@FreeBSD.ORG Fri Oct 7 01:12:38 2011 Return-Path: Delivered-To: freebsd-net@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 6DBAE106564A; Fri, 7 Oct 2011 01:12:38 +0000 (UTC) (envelope-from s.dave.jones@gmail.com) Received: from mail-vx0-f182.google.com (mail-vx0-f182.google.com [209.85.220.182]) by mx1.freebsd.org (Postfix) with ESMTP id B642A8FC08; Fri, 7 Oct 2011 01:12:37 +0000 (UTC) Received: by vcbf13 with SMTP id f13so3879365vcb.13 for ; Thu, 06 Oct 2011 18:12:36 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; bh=4hXn8v5lEBADsBNYiWgRZmkTl+ShK0zrMoxepMpR9zs=; b=LsoiAKW2GPeaaaWeI38YTc59eYbBPNnaj9u1Ih1A9guFBsXUA6g6LP9C9Z7VxL4u87 7M4GzFHyrvFbwvE214w35TiHyfNZmXhXqAMDDU1jFma93XmePhQoFBgfN2xOhYfa40qr SzKkRchroOKpYOSh0M8i1Q9xcvq5172NLD2KQ= MIME-Version: 1.0 Received: by 10.52.31.41 with SMTP id x9mr449285vdh.42.1317949956922; Thu, 06 Oct 2011 18:12:36 -0700 (PDT) Received: by 10.52.184.135 with HTTP; Thu, 6 Oct 2011 18:12:36 -0700 (PDT) In-Reply-To: <86y5x0ooik.fsf@in138.ua3> References: <8662kcigif.fsf@kopusha.home.net> <86y5x0ooik.fsf@in138.ua3> Date: Fri, 7 Oct 2011 09:12:36 +0800 Message-ID: From: dave jones To: Mikolaj Golub Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Cc: "freebsd-net@freebsd.org" , Adrian Chadd , Robert Watson , "K. Macy" , Arnaud Lacombe Subject: Re: Kernel panic on FreeBSD 9.0-beta2 X-BeenThere: freebsd-net@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: Networking and TCP/IP with FreeBSD List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 07 Oct 2011 01:12:38 -0000 2011/10/4 Mikolaj Golub : > > On Sat, 1 Oct 2011 14:15:45 +0800 dave jones wrote: > > =A0dj> On Fri, Sep 30, 2011 at 9:41 PM, Robert Watson wrote: > =A0>> > =A0>> On Wed, 28 Sep 2011, Mikolaj Golub wrote: > =A0>> > =A0>>> On Mon, 26 Sep 2011 16:12:55 +0200 K. Macy wrote: > =A0>>> > =A0>>> KM> Sorry, didn't look at the images (limited bw), I've seen somet= hing KM> > =A0>>> like this before in timewait. This "can't happen" with UDP so will= be KM> > =A0>>> interested in learning more about the bug. > =A0>>> > =A0>>> The panic can be easily triggered by this: > =A0>> > =A0>> Hi: > =A0>> > =A0>> Just catching up on this thread. =A0I think the analysis here is ge= nerally > =A0>> right: in 9.0, you're much more likely to see an inpcb with its in_= socket > =A0>> pointer cleared in the hash list than in prior releases, and > =A0>> in_pcbbind_setup() trips over this. > =A0>> > =A0>> However, at least on first glance (and from the perspective of inva= riants > =A0>> here), I think the bug is actualy that in_pcbbind_setup() is asking > =A0>> in_pcblookup_local() for an inpcb and then access the returned inpc= b's > =A0>> in_socket pointer without acquiring a lock on the inpcb. =A0Structu= rally, it > =A0>> can't acquire this lock for lock order reasons -- it already holds = the lock > =A0>> on its own inpcb. =A0Therefore, we should only access fields that a= re safe to > =A0>> follow in an inpcb when you hold a reference via the hash lock and = not a > =A0>> lock on the inpcb itself, which appears generally OK (+/-) for all = the > =A0>> fields in that clause but the t->inp_socket->so_options dereference= . > =A0>> > =A0>> A preferred fix would cache the SO_REUSEPORT flag in an inpcb-layer= field, > =A0>> such as inp_flags2, giving us access to its value without having to= walk > =A0>> into the attached (or not) socket. > =A0>> > =A0>> This raises another structural question, which is whether we need a= new > =A0>> inp_foo flags field that is protected explicitly by the hash lock, = and not > =A0>> by the inpcb lock, which could hold fields relevant to address bind= ing. =A0I > =A0>> don't think we need to solve that problem in this context, as a sli= ghtly > =A0>> race on SO_REUSEPORT is likely acceptable. > =A0>> > =A0>> The suggested fix does perform the desired function of explicitly d= etaching > =A0>> the inpcb from the hash list before the socket is disconnected from= the > =A0>> inpcb. However, it's incomplete in that the invariant that's being = broken is > =A0>> also relied on for other protocols (such as raw sockets). =A0The co= rrect > =A0>> invariant is that inp_socket is safe to follow unconditionally if a= n inpcb > =A0>> is locked and INP_DROPPED isn't set -- the bug is in "locked" not i= n > =A0>> "INP_DROPPED", which is why I think this is the wrong fix, even tho= ugh it > =A0>> prevents a panic :-). > > =A0dj> Hello Robert, > > =A0dj> Thank you for taking your valuable time to find out the problem. > =A0dj> Since I don't have idea about network internals, would you have a = patch > =A0dj> about this? I'd be glad to test it, thanks again. > > Here is the patch that implements what Robert suggests. > > Dave, could you test it? Sure. Thanks for cooking the patch. Machines have been running two days now without panic. > -- > Mikolaj Golub Best regards, Dave.