From: Harald Welte Subject: Re: Linux kernel crashes when incoming/outgoing interfaces differ Date: Fri, 17 Jan 2003 01:10:44 +0100 Sender: netfilter-devel-admin@lists.netfilter.org Message-ID: <20030117001044.GL7774@sunbeam.de.gnumonks.org> References: <20030115130419.GX7242%bowles@ambisys.com> Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="/unnNtmY43mpUSKx" Cc: netfilter-devel@lists.netfilter.org Return-path: To: "Eric E. Bowles" Content-Disposition: inline In-Reply-To: <20030115130419.GX7242%bowles@ambisys.com> Errors-To: netfilter-devel-admin@lists.netfilter.org List-Help: List-Post: List-Subscribe: , List-Unsubscribe: , List-Archive: List-Id: netfilter-devel.vger.kernel.org --/unnNtmY43mpUSKx Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Wed, Jan 15, 2003 at 10:04:19PM +0900, Eric E. Bowles wrote: > [I originally sent this to the OpenVPN devel mailing list, and it > was suggested that I forward it to other lists since it might be a > problem involving netfilter and/or the kernel, so here goes....] >=20 > Hi there, >=20 > I've been experiencing kernel crashes with OpenVPN (http://openvpn.sf.net= )=20 > on Linux under a certain set of conditions, and wondered if anybody else = can=20 > reproduce this. Hi! This bug report sounds very interesting. I have put it into our brand-new bugzilla system, you can keep track of the bug at=20 https://bugzilla.netfilter.org/cgi-bin/bugzilla/show_bug.cgi?id=3D22 > Here are the steps needed to cause the crash: [...] I'm trying to build a test setup for this case ASAP, I think I'll have some time at the weekend. =20 > What I think is happening is this: when an incoming connection arrives > on Host-A's TUN interface, the REJECT rule is triggered, causing an=20 > "ICMP unreachable" to be sent back. However, since there is no return=20 > route back across the tunnel, the default route is selected, which causes= =20 > the ICMP packet to be < incoming interface>> (i.e., the tunnel). Since the incoming (tun) and=20 > outgoing (eth) interfaces have different link layer header sizes, the=20 > kernel is crashing in skb_push(). Sounds like a plausible explanation. As stated above, as soon as I can reproduce it, I'll do further debugging and hopefully find a fix for this issue. > Thanks, > --eric Thanks for reporting this bug. --=20 - Harald Welte / laforge@gnumonks.org http://www.gnumonks.org/ =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D "If this were a dictatorship, it'd be a heck of a lot easier, just so long as I'm the dictator." -- George W. Bush Dec 18, 2000 --/unnNtmY43mpUSKx Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.0.6 (GNU/Linux) Comment: For info see http://www.gnupg.org iD8DBQE+J0oEXaXGVTD0i/8RAv7GAJ9C7eAQkCg7GaxJwG0GNAq344rpGQCcDQ3g 2iZ3dXjWnMfDEq77q7lE2Bo= =mpBC -----END PGP SIGNATURE----- --/unnNtmY43mpUSKx--