Return-Path: Content-Type: text/plain; charset=windows-1252 Mime-Version: 1.0 (Mac OS X Mail 8.2 \(2070.6\)) Subject: Re: DBus API for LTK OOB keys From: Marcel Holtmann In-Reply-To: <54F8B881.4050506@ubnt.com> Date: Thu, 5 Mar 2015 14:08:39 -0800 Cc: "linux-bluetooth@vger.kernel.org" Message-Id: <280B8CCC-6E98-477A-B805-D91966220207@holtmann.org> References: <54F8584D.6020301@ubnt.com> <20150305134024.GA20862@t440s> <54F862A3.80400@ubnt.com> <8EB5FD2F-4C46-41F8-95BA-4ACBEF0BCDE4@holtmann.org> <54F8B881.4050506@ubnt.com> To: Andrejs Hanins Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Andrejs, >>>>> There is a MGMT "Load Long Term Keys Command" to feed keys to the Kernel >>>>> which are stored in the BlueZ settings storage file and read during adapter >>>>> init (load_devices->load_ltks). I searched through the code and couldn't >>>>> find any means to feed LTK keys from "the outside", for example through DBus >>>>> API. This is needed for an LE OOB pairing scheme, when key is known in >>>>> advance by both parties and is not derived from pairing procedure. Is there >>>>> a standardized way to add LTK keys "manually" or this is not supported-yet >>>>> feature? According to setting storage rules "Direct access to the storage >>>>> outside from bluetoothd is highly discouraged". >>>> Are the keys provisioned beforehand or is this something that can happen >>>> at any time when bluetoothd is already running? If it's the former then >>>> a custom bluetoothd plugin that gives bluetoothd core an extra set of >>>> keys could be one way to go. If it's the latter, then things get tricky >>>> since the mgmt command wipes all existing keys away before adding new >>>> ones. >>> In my particular scenario, the OOB key is provisioned only once and beforehand and used to connect with multiple LE devices. LE devices get this key via some proprietary mechanism. So it is kind of "global master key". As such, it is not a problem to restart the BlueZ daemon after the key is (re-)provisioned. >> using the same LTK for multi devices is a really bad idea. This is not how Bluetooth LE security was designed. My advise would be strictly against doing that in any product. > Beg your pardon, I mixed up OOB and LTK, my bad. The proper question is to how to feed OOB Datafor OOB pairing, which in turn used to generate STK and then unique per-device LTK as you have described below. Actually, I have found a way - btd_adapter_add_remote_oob_data() which is exactly what I need. But, in order for OOB Pairing to be started, the "SMP Pairing Request (0x01)" should indicate the presence of OOB Data. In kernel 3.18.6 it does not happening simply because the macro SMP_OOB_PRESENT is not used at all. In latest kernel 3.19 some changes were made in regard to OOB Data and macro is now used (see build_pairing_cmd) but only if local device and authreq from remote party both support "Secure Connections" which is in turn BT Core 4.2 feature. But OOB pairing is supported also in BT Core 4.0, isn't it? So my question boils down to the following (it is all about LE bearers only): > 1. Is LE OOB Pairing not supported before kernel 3.19 ? (see SMP_OOB_PRESENT macro which is not used) > 2. Is LE OOB Pairing is still broken in 3.19, because it works only when both sides support "Secure Connections" thus are 4.2 version devices? we still have a problem with LE OOB pairing in the sense of giving the kernel enough information. That will be fixed hopefully pretty soon. The proposal on how to do that is in mgmt-api.txt. I think Johan is working on it. The main problem is that we can currently only get OOB information for BR/EDR side of things. The missing part is to the get the LE OOB information. And of course there is a difference between LE Legacy Pairing and LE Secure Connections. >>>> If your OOB scheme is this latter "non-pre-provisioned" one I'd wonder >>>> why you're not using the standard OOB mechanism provided by LE SMP, >>>> since for that we do have at least partial support. >>> I'm now confused a bit. Indeed, I want to use OOB mechanism provided by LE SMP, but in order to start OOB pairing, the OOB key itself should be known to both sides (central and peripheral). For the peripheral I have my own ways to do it (proprietary), but the main question it how to give the key value to the central which is BlueZ in this case. >> When using LE OOB pairing (Legacy Pairing and Secure Connections), then at least you get a sense of key strength that is guaranteed based on how LE security is designed. So doing LE OOB pairing is a good idea. You feed different information into OOB pairing. You share and OOB secret between two devices and based on that they can pair. However they need to be in range to actually use that OOB secret to pair. Pairing requires a connection. The advantage with pairing is that you get a proper key with a proper strength. So LE Legacy Pairing needs the Security Manager TK value and LE Secure Connections needs the Confirmation and Random values. However right now, we can not get these ones from the kernel. As a side note, I added tools/oobtest utility where you can test this between two controllers attached to the same host. Regards Marcel