Return-Path: MIME-Version: 1.0 In-Reply-To: References: <1457094267-5925-1-git-send-email-luiz.dentz@gmail.com> <1457094267-5925-2-git-send-email-luiz.dentz@gmail.com> Date: Thu, 10 Mar 2016 10:30:45 +0200 Message-ID: Subject: Re: [PATCH BlueZ 2/2] core/device: Don't set MTU when acting as a peripheral From: Luiz Augusto von Dentz To: =?UTF-8?Q?=C5=81ukasz_Rymanowski?= Cc: "linux-bluetooth@vger.kernel.org" Content-Type: text/plain; charset=UTF-8 Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Lukasz, On Thu, Mar 10, 2016 at 9:59 AM, Łukasz Rymanowski wrote: > Hi Luiz, > > On 9 March 2016 at 13:13, Luiz Augusto von Dentz wrote: >> Hi Lukasz, >> >> On Fri, Mar 4, 2016 at 5:46 PM, Łukasz Rymanowski >> wrote: >>> Hi Luiz, >>> >>> On 4 March 2016 at 14:12, Luiz Augusto von Dentz wrote: >>>> Hi Łukasz, >>>> >>>> On Fri, Mar 4, 2016 at 2:51 PM, Łukasz Rymanowski >>>> wrote: >>>>> Hi Luiz, >>>>> >>>>> On 4 March 2016 at 13:24, Luiz Augusto von Dentz wrote: >>>>>> From: Luiz Augusto von Dentz >>>>>> >>>>>> Some Android versions don't seem to cope well with the fact that the >>>>>> peripheral can actually set a MTU thus leave the MTU as the default. >>>>> >>>>> Just tested that and for me connection with Android works OK even BlueZ >>>>> started Exchange MTU procedure. Tested on Android 5 and 6. >>>>> >>>>> I saw once the issue that Android could not retrieve services after >>>>> connection to BlueZ >>>>> but from the logs it looked like BlueZ start search for primary >>>>> services and Android not. >>>>> If Android make to sent search request for primary services first then >>>>> it was fine. >>>>> >>>>> Maybe Android don't like BlueZ to start search primary if Android is >>>>> initiator? But this is their bug >>>>> BTW. I saw it on Android 5 not yet on 6. >>>> >>>> With 4.4 it actually made a difference. >>>> >>>>> >>>>>> --- >>>>>> src/attrib-server.c | 2 +- >>>>>> src/device.c | 8 +++++--- >>>>>> src/device.h | 2 +- >>>>>> src/gatt-database.c | 2 +- >>>>>> 4 files changed, 8 insertions(+), 6 deletions(-) >>>>>> >>>>>> diff --git a/src/attrib-server.c b/src/attrib-server.c >>>>>> index 4439c27..e92ca5c 100644 >>>>>> --- a/src/attrib-server.c >>>>>> +++ b/src/attrib-server.c >>>>>> @@ -1281,7 +1281,7 @@ static void connect_event(GIOChannel *io, GError *gerr, void *user_data) >>>>>> if (!device) >>>>>> return; >>>>>> >>>>>> - device_attach_att(device, io); >>>>>> + device_attach_att(device, io, false); >>>>>> } >>>>>> >>>>>> static gboolean register_core_services(struct gatt_server *server) >>>>>> diff --git a/src/device.c b/src/device.c >>>>>> index 14e850e..b4bc593 100644 >>>>>> --- a/src/device.c >>>>>> +++ b/src/device.c >>>>>> @@ -4664,7 +4664,7 @@ static bool remote_counter(uint32_t *sign_cnt, void *user_data) >>>>>> return true; >>>>>> } >>>>>> >>>>>> -bool device_attach_att(struct btd_device *dev, GIOChannel *io) >>>>>> +bool device_attach_att(struct btd_device *dev, GIOChannel *io, bool initiator) >>>>>> { >>>>>> GError *gerr = NULL; >>>>>> GAttrib *attrib; >>>>>> @@ -4699,7 +4699,9 @@ bool device_attach_att(struct btd_device *dev, GIOChannel *io) >>>>>> } >>>>>> } >>>>>> >>>>>> - dev->att_mtu = MIN(mtu, BT_ATT_MAX_LE_MTU); >>>>>> + /* Only attempt to set MTU if initiator/central */ >>>>> >>>>> Keep in mind that MTU echange procedure is bonded to Client/Server >>>>> role not central/peripheral. >>>> >>>> Well actually is not either GAP or GATT role related since both side >>>> are in fact allowed to do that, so this is just a workaround. >>> >>> According to spec Exchange MTU procedure can be started by Client or >>> devices supporting Client/Server role. >>> BlueZ always acts as client/server this is why you can always start >>> this procedure. >>> It is really not related to central/peripheral. >> >> Well the working of client/server role seems very odd to say the >> least, why the use of client and server role if they could always be >> both, but it doesn't change the fact that either side can send MTU >> Exchange. >> > > Server only device cannot start Exchange MTU procedure. But since BlueZ cannot > be configured this way that is not a case. Not only we cannot be configured as server, but even if we do the roles are only relevant for PTS because otherwise there is no way to tell the roles of the remote device. >>>> >>>>> >>>>>> + dev->att_mtu = initiator ? MIN(mtu, BT_ATT_MAX_LE_MTU) : >>>>>> + BT_ATT_DEFAULT_LE_MTU; >>>>>> attrib = g_attrib_new(io, dev->att_mtu, false); >>>>>> if (!attrib) { >>>>>> error("Unable to create new GAttrib instance"); >>>>>> @@ -4768,7 +4770,7 @@ static void att_connect_cb(GIOChannel *io, GError *gerr, gpointer user_data) >>>>>> goto done; >>>>>> } >>>>>> >>>>>> - if (!device_attach_att(device, io)) >>>>>> + if (!device_attach_att(device, io, true)) >>>>>> goto done; >>>>>> >>>>>> if (attcb->success) >>>>>> diff --git a/src/device.h b/src/device.h >>>>>> index db10827..5c01ed0 100644 >>>>>> --- a/src/device.h >>>>>> +++ b/src/device.h >>>>>> @@ -72,7 +72,7 @@ struct bt_gatt_client *btd_device_get_gatt_client(struct btd_device *device); >>>>>> struct bt_gatt_server *btd_device_get_gatt_server(struct btd_device *device); >>>>>> void btd_device_gatt_set_service_changed(struct btd_device *device, >>>>>> uint16_t start, uint16_t end); >>>>>> -bool device_attach_att(struct btd_device *dev, GIOChannel *io); >>>>>> +bool device_attach_att(struct btd_device *dev, GIOChannel *io, bool initiator); >>>>>> void btd_device_add_uuid(struct btd_device *device, const char *uuid); >>>>>> void device_add_eir_uuids(struct btd_device *dev, GSList *uuids); >>>>>> void device_set_manufacturer_data(struct btd_device *dev, GSList *list); >>>>>> diff --git a/src/gatt-database.c b/src/gatt-database.c >>>>>> index d906e2b..4df4a16 100644 >>>>>> --- a/src/gatt-database.c >>>>>> +++ b/src/gatt-database.c >>>>>> @@ -499,7 +499,7 @@ static void connect_cb(GIOChannel *io, GError *gerr, gpointer user_data) >>>>>> if (!device) >>>>>> return; >>>>>> >>>>>> - device_attach_att(device, io); >>>>>> + device_attach_att(device, io, false); >>>>> >>>>> Not sure if it works when you use white list for autoconnect because >>>>> iirc we got this callback when BlueZ is initiator >>>>> then. But still initiator or not, the important thing is if we act as >>>>> client or server. >>>> >>>> Probably not, >>> >>> I just run the test: >>> >>> 1) add device to autoconnect list. >>> 2) Simulate disconnect by power off. >>> 3) After then power on device and wait until auto connect do its job: >>> >>> I got this: >>> >>> Mar 4 16:30:36 t450s bluetoothd[31211]: >>> src/adapter.c:connected_callback() hci0 device XX:XX:XX:XX:XX:XX >>> connected eir_len 14 >>> Mar 4 16:30:36 t450s bluetoothd[31211]: >>> src/gatt-database.c:connect_cb() New incoming LE ATT connection >>> >>> That means that with this patch after reconnect a default MTU will be >>> used unless remote device starts >>> exchange MTU procedure. >>> >>> I will have to add more logic to detect how it got >>>> connected, maybe we can actually check who is the master so that the >>>> master should be the one controlling the MTU. >>>> >>> >>> Maybe we can do exchange MTU later after discovery ? For discovery >>> default MTU should be enough. >> >> Well we still need to detect the role to do that anyway, and I >> wouldn't bother with MTU request being the slave since the MTU shall >> be symmetric: >> >> BLUETOOTH SPECIFICATION Version 4.2 [Vol 3, Part F] page 484: >> 1. A device's Exchange MTU Request shall contain the same MTU as the >> device's Exchange MTU Response (i.e. the MTU shall be symmetric). > > > Yes, MTU is symmetric and this is not a discussion point. > Bottom line here is like this: > > 1. Current BlueZ acts correctly in MTU exchange context > 2. We try to workaround for not correctly working Android 4.4 which > has issue with BlueZ sending Exchange MTU request Im fine dropping this if it only affects 4.4, but we need to check later version if that was really fixed. > 3. role on the link is not relevant. There can be devices being > Peripheral/Clienr or Central/Server. Central/Server will never send > Exchange MTU Request and if BlueZ rely on the role on the link we will > operate on Default MTU even though we can do better. Indeed, different profiles may have different roles, I don't think I have anything otherwise. Anyway most devices respond with Default MTU anyway so chances are this doesn't change anything. -- Luiz Augusto von Dentz