From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:48371) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1eZyvI-0003Pt-OM for qemu-devel@nongnu.org; Fri, 12 Jan 2018 07:59:08 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1eZyvH-0002sj-4x for qemu-devel@nongnu.org; Fri, 12 Jan 2018 07:59:04 -0500 Received: from mx1.redhat.com ([209.132.183.28]:51466) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1eZyvG-0002qX-QH for qemu-devel@nongnu.org; Fri, 12 Jan 2018 07:59:03 -0500 Received: from smtp.corp.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.15]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id C444167729 for ; Fri, 12 Jan 2018 12:59:01 +0000 (UTC) From: Gerd Hoffmann Date: Fri, 12 Jan 2018 13:58:52 +0100 Message-Id: <20180112125854.18261-13-kraxel@redhat.com> In-Reply-To: <20180112125854.18261-1-kraxel@redhat.com> References: <20180112125854.18261-1-kraxel@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Subject: [Qemu-devel] [PULL 12/14] ui: place a hard cap on VNC server output buffer size List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: "Daniel P. Berrange" , Gerd Hoffmann From: "Daniel P. Berrange" The previous patches fix problems with throttling of forced framebuffer u= pdates and audio data capture that would cause the QEMU output buffer size to gr= ow without bound. Those fixes are graceful in that once the client catches u= p with reading data from the server, everything continues operating normally. There is some data which the server sends to the client that is impractic= al to throttle. Specifically there are various pseudo framebuffer update encodi= ngs to inform the client of things like desktop resizes, pointer changes, audio playback start/stop, LED state and so on. These generally only involve se= nding a very small amount of data to the client, but a malicious guest might be= able to do things that trigger these changes at a very high rate. Throttling t= hem is not practical as missed or delayed events would cause broken behaviour fo= r the client. This patch thus takes a more forceful approach of setting an absolute upp= er bound on the amount of data we permit to be present in the output buffer = at any time. The previous patch set a threshold for throttling the output bu= ffer by allowing an amount of data equivalent to one complete framebuffer upda= te and one seconds worth of audio data. On top of this it allowed for one furthe= r forced framebuffer update to be queued. To be conservative, we thus take that throttling threshold and multiply i= t by 5 to form an absolute upper bound. If this bound is hit during vnc_write(= ) we forceably disconnect the client, refusing to queue further data. This lim= it is high enough that it should never be hit unless a malicious client is tryi= ng to exploit the sever, or the network is completely saturated preventing any = sending of data on the socket. This completes the fix for CVE-2017-15124 started in the previous patches= . Signed-off-by: Daniel P. Berrange Reviewed-by: Darren Kenny Reviewed-by: Marc-Andr=C3=A9 Lureau Message-id: 20171218191228.31018-12-berrange@redhat.com Signed-off-by: Gerd Hoffmann --- ui/vnc.c | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/ui/vnc.c b/ui/vnc.c index 4805ac41d0..e53e84587a 100644 --- a/ui/vnc.c +++ b/ui/vnc.c @@ -1521,8 +1521,37 @@ gboolean vnc_client_io(QIOChannel *ioc G_GNUC_UNUS= ED, } =20 =20 +/* + * Scale factor to apply to vs->throttle_output_offset when checking for + * hard limit. Worst case normal usage could be x2, if we have a complet= e + * incremental update and complete forced update in the output buffer. + * So x3 should be good enough, but we pick x5 to be conservative and th= us + * (hopefully) never trigger incorrectly. + */ +#define VNC_THROTTLE_OUTPUT_LIMIT_SCALE 5 + void vnc_write(VncState *vs, const void *data, size_t len) { + if (vs->disconnecting) { + return; + } + /* Protection against malicious client/guest to prevent our output + * buffer growing without bound if client stops reading data. This + * should rarely trigger, because we have earlier throttling code + * which stops issuing framebuffer updates and drops audio data + * if the throttle_output_offset value is exceeded. So we only reach + * this higher level if a huge number of pseudo-encodings get + * triggered while data can't be sent on the socket. + * + * NB throttle_output_offset can be zero during early protocol + * handshake, or from the job thread's VncState clone + */ + if (vs->throttle_output_offset !=3D 0 && + vs->output.offset > (vs->throttle_output_offset * + VNC_THROTTLE_OUTPUT_LIMIT_SCALE)) { + vnc_disconnect_start(vs); + return; + } buffer_reserve(&vs->output, len); =20 if (vs->ioc !=3D NULL && buffer_empty(&vs->output)) { --=20 2.9.3