Received-SPF: pass (google.com: domain of linux-bluetooth-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <CANFkB1ydGuwKRjusy3U0HCQo8E88XbZTh_UdPZ8wujt7P5KKnQ@mail.gmail.com>
 <CABBYNZLjV3PQ7Hm3PJZcTcs64aLzWLBNfooSYuZmSywv0RJG0g@mail.gmail.com>
 <CANFkB1yr3N_Twac=ocxBRst79-gmdu8=6VTD6oPgdGT0rb_HaQ@mail.gmail.com>
 <CANFkB1xEesP1cMmK4VGi5cH-v80ho00wT24q=+4QQSdyPegPQQ@mail.gmail.com>
 <CABBYNZKSsyqfiWYjpuZNYtKAZb4C2q73YwOj=KoUFq71gLs7Hg@mail.gmail.com>
 <CANFkB1zHTZF3Mcgn5GPBA0km+BNyL3bb9=MJFuSNFe6x5zgimQ@mail.gmail.com>
 <CABBYNZLWN3+51F-sNJ5yPc2bw+x4Ep6rCwcFLn0bfrwArL=Hyw@mail.gmail.com>
 <CABBYNZL-fjTd8JdmRRM5Y26d41KwDxvvuaED+j2+yPT-RBpMtw@mail.gmail.com>
 <CANFkB1w2UUq7W0EqEdskebhKcEtkZ-33q-Co5VF_WKf7dHsJtg@mail.gmail.com>
 <CANFkB1zMOYTD8ZsDHczHqce9ipgyu9-fEeXEZqTzuf=T4UufLQ@mail.gmail.com>
 <CANFkB1yOadrpibp4ZNE8e0E+-Wy4z_FV3LC=B7HZzUjGAh3DVA@mail.gmail.com> <CABBYNZK_CdiXaZzw++x8fzarP_WhCehLia17EpruVTduKZ=Png@mail.gmail.com>
In-Reply-To: <CABBYNZK_CdiXaZzw++x8fzarP_WhCehLia17EpruVTduKZ=Png@mail.gmail.com>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Wed, 23 Mar 2022 12:52:59 -0700
Message-ID: <CABBYNZK7WWVQ3kof+fYZV4VkJrxQe9wkf9RDsHojPEcS7FLFsQ@mail.gmail.com>
Subject: Re: GATT issue, possible bluez bug?
To:     Adam Pigg <adam@piggz.co.uk>
Cc:     "linux-bluetooth@vger.kernel.org" <linux-bluetooth@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Adam,

On Wed, Mar 23, 2022 at 12:45 PM Luiz Augusto von Dentz
<luiz.dentz@gmail.com> wrote:
>
> Hi Adam,
>
> On Wed, Mar 23, 2022 at 7:46 AM Adam Pigg <adam@piggz.co.uk> wrote:
> >
> > On Wed, 23 Mar 2022 at 13:36, Adam Pigg <adam@piggz.co.uk> wrote:
> > >
> > > Hi Luiz
> > >
> > > On Wed, 23 Mar 2022 at 09:11, Adam Pigg <adam@piggz.co.uk> wrote:
> > > >
> > > > Hi Luiz
> > > >
> > > > Good that you clearly know more about this than me!...
> > > >
> > > > On Tue, 22 Mar 2022 at 23:44, Luiz Augusto von Dentz
> > > > <luiz.dentz@gmail.com> wrote:
> > > > >
> > > > > Hi Adam,
> > > > >
> > > > > On Tue, Mar 22, 2022 at 4:35 PM Luiz Augusto von Dentz
> > > > > <luiz.dentz@gmail.com> wrote:
> > > > > >
> > > > > > Hi Adam,
> > > > > >
> > > > > > On Tue, Mar 22, 2022 at 4:09 PM Adam Pigg <adam@piggz.co.uk> wrote:
> > > > > > >
> > > > > > > Thanks Luiz
> > > > > > >
> > > > > > > On Tue, 22 Mar 2022 at 21:35, Luiz Augusto von Dentz
> > > > > > > <luiz.dentz@gmail.com> wrote:
> > > > > > > >
> > > > > > > > Hi Adam,
> > > > > > > >
> > > > > > > > On Tue, Mar 22, 2022 at 3:40 AM Adam Pigg <adam@piggz.co.uk> wrote:
> > > > > > > > >
> > > > > > > > > Hi Luiz
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Tue, 22 Mar 2022 at 07:55, Adam Pigg <adam@piggz.co.uk> wrote:
> > > > > > > > > >
> > > > > > > > > > Hi Luiz
> > > > > > > > > >
> > > > > > > > > > On Tue, 22 Mar 2022 at 00:44, Luiz Augusto von Dentz
> > > > > > > > > > <luiz.dentz@gmail.com> wrote:
> > > > > > > > > > >
> > > > > > > > > > > Hi Adam,
> > > > > > > > > > >
> > > > > > > > > > > On Mon, Mar 21, 2022 at 4:03 PM Adam Pigg <adam@piggz.co.uk> wrote:
> > > > > > > > > > > >
> > > > > > > > > > > > Hi
> > > > > > > > > > > >
> > > > > > > > > > > > A bit of background, I write a linux desktop/mobile app called
> > > > > > > > > > > > Amazfish which interfaces with several watches over BLE using their
> > > > > > > > > > > > GATT services.  To do this, I use the bluez dbus api (technically a
> > > > > > > > > > > > thin wrapper around it I wrote called qble
> > > > > > > > > > > > https://github.com/piggz/qble)
> > > > > > > > > > > >
> > > > > > > > > > > > All has been good so far, I support several generations of
> > > > > > > > > > > > Huami/Amazfit watches, as well as the open source Pinetime and
> > > > > > > > > > > > Bangle.js.  For the Amazfit watches, i have implementations for older
> > > > > > > > > > > > devies such as the Bip, and newer ones such as the GTS.
> > > > > > > > > > > >
> > > > > > > > > > > > Much of the reverse engineering comes from the Android Gadget Bridge
> > > > > > > > > > > > project, which supports many more devices.
> > > > > > > > > > > >
> > > > > > > > > > > > My community of users donated to buy me a newer device called a GTR2,
> > > > > > > > > > > > which, according to the GB devs uses the same protocol as the slightly
> > > > > > > > > > > > older GTS, and the packet captures I have from Android would support
> > > > > > > > > > > > this.
> > > > > > > > > > > >
> > > > > > > > > > > > But this is where my trouble starts with Bluez, my existing
> > > > > > > > > > > > implementation doesnt work at all.  Normally, after a connection, I
> > > > > > > > > > > > would wait for the ServicesResolved signal, which happens pretty fast
> > > > > > > > > > > > on all other devices, but on the GTR2, it takes about 30 seconds, by
> > > > > > > > > > > > which time, the watch has disconnected. (i get a disconnected signal
> > > > > > > > > > > > immediately after the ServicesResolved signal)
> > > > > > > > > > > >
> > > > > > > > > > > > To rule out my code, I have tried several things:
> > > > > > > > > > > > Gatttool
> > > > > > > > > > > > With gattool, i can connect, get the services, enable a notification,
> > > > > > > > > > > > write a value and get the expected results seemingly fine
> > > > > > > > > > > >
> > > > > > > > > > > > Python-gatt (using the bluez dbus api)
> > > > > > > > > > > > Im unable to iterate the services, like my app, it takes 30 seconds to
> > > > > > > > > > > > get the signal and then swiftly disconnects
> > > > > > > > > > > >
> > > > > > > > > > > > Gattlib (https://github.com/labapart/gattlib)
> > > > > > > > > > > > Gattlib is interesting as it appears to have "borrowed" much of its
> > > > > > > > > > > > code directly from bluez.  When built against the system bluez, if the
> > > > > > > > > > > > version is > 5.42, it will use the dbus api.  When I do this, again im
> > > > > > > > > > > > unable to list services on the watch.  However, if I edit the build to
> > > > > > > > > > > > force it to use its internal gatt implementation, which appears to be
> > > > > > > > > > > > the same one used by gatttool, then, it IS able to interrogate the
> > > > > > > > > > > > watch.
> > > > > > > > > > > >
> > > > > > > > > > > > I have attached 3 files
> > > > > > > > > > > > 1. test python program which should print services, and associated btmon
> > > > > > > > > > > > 2. btmon output while using gatttool
> > > > > > > > > > > > 3. btmon output running gattlib discover example
> > > > > > > > > > > >
> > > > > > > > > > > > Note, other than discovery, I havnt been able to get gattlib to
> > > > > > > > > > > > read/write/notify!
> > > > > > > > > > > >
> > > > > > > > > > > > It seems as though I may be triggering a bug in the bluez dbus api?
> > > > > > > > > > > > Can anyone suggest anything?
> > > > > > > > > > >
> > > > > > > > > > > What version are you using? I would first try with the latest to see
> > > > > > > > > > > if that something already fixed.
> > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > Im using 5.63 already, which is the latest tag (on opensuse tumbleweed)
> > > > > > > > > >
> > > > > > > > > > I'll also try bluetoothctl and attach logs using that, though I
> > > > > > > > > > suspect it will behave the same as the python library and my own
> > > > > > > > > > library.
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > > Attached is the output of btmon while using bluetoothctl
> > > > > > > > >
> > > > > > > > > bluetoothctl behaves the same as my lib and python, it is unable to
> > > > > > > > > list services, and after a while, emits the ServicesResolved and
> > > > > > > > > disconnected messages at the same time.  Im sure it should be able to
> > > > > > > > > list the services immediately after connect just like gatttool can.
> > > > > > > >
> > > > > > > > Looks like there is multiple rounds of MTU Exchange, not sure if that
> > > > > > > > is actually a problem but the spec does say it shall only be sent once
> > > > > > > > per client:
> > > > > > > >
> > > > > > > > BLUETOOTH CORE SPECIFICATION Version 5.3 | Vol 3, Part F
> > > > > > > > page 1424:
> > > > > > > >
> > > > > > > > This request shall only be sent once during a connection by the client.
> > > > > > > >
> > > > > > > > There is also no response to frame #30:
> > > > > > > > < ACL Data TX: Handle 3585 flags 0x00 dlen 7
> > > > > > > >
> > > > > > > >                            #30 [hci0] 26.641557
> > > > > > > >       ATT: Read Request (0x0a) len 2
> > > > > > > >         Handle: 0x0018
> > > > > > > >
> > > > > > > > So after 30 seconds (ATT timeout) without receiving any response it disconnects:
> > > > > > > >
> > > > > > >
> > > > > > > So, the verdict is that the watch isnt acting to spec right?
> > > > > > >
> > > > > > > Why does gatttool work ok, is it a completely different implementation?
> > > > > > >
> > > > > > > I understand if wouldnt be to spec, but would it be possible to make
> > > > > > > bluez more resilient to devices not acting to spec?  Android and iOS
> > > > > > > work just fine in this regard, so could bluez handle mis-behaving
> > > > > > > devices more flexibly?
> > > > > >
> > > > > > That I can't really explain, perhaps it is timing related, something
> > > > > > with Exchange MTU since it appears to stop the device from responding
> > > > > > when it happens a second time.
> > > > > >
> > > > > > > Do you have any suggestions for things I could try? I looked for
> > > > > > > similar lines in the working gatttool log, and it doesnt have anything
> > > > > > > like that, seems to send flags 0x00 dlen 9 (instead of 7?)  ... how is
> > > > > > > it behaving differently?
> > > > > >
> > > > > > Not sure really, does Android/iOS Exchange the MTU? Maybe we need a
> > > > > > timer to start it later or perhaps the problem first response to
> > > > > > exchange MTU does actually use the final_mtu which makes the watch to
> > > > > > trigger yet another exchange to have both rx and tx MTU the same so
> > > > > > try with the following change:
> > > > > >
> > > > > > diff --git a/src/shared/gatt-server.c b/src/shared/gatt-server.c
> > > > > > index 2adb4afbf..d326782bf 100644
> > > > > > --- a/src/shared/gatt-server.c
> > > > > > +++ b/src/shared/gatt-server.c
> > > > > > @@ -1499,7 +1499,7 @@ static void exchange_mtu_cb(struct bt_att_chan
> > > > > > *chan, uint8_t opcode,
> > > > > >         final_mtu = MAX(MIN(client_rx_mtu, server->mtu), BT_ATT_DEFAULT_LE_MTU);
> > > > > >
> > > > > >         /* Respond with the server MTU */
> > > > > > -       put_le16(server->mtu, rsp_pdu);
> > > > > > +       put_le16(final_mtu, rsp_pdu);
> > > > > >         bt_att_chan_send_rsp(chan, BT_ATT_OP_MTU_RSP, rsp_pdu, 2);
> > > > > >
> > > > > >         /* Set MTU to be the minimum */
> > > > >
> > > > > Hmm Im not sure if this will work since the peripheral seems to
> > > > > respond different values 252 vs 247, also the spec does say that:
> > > > >
> > > > > 'The Server Rx MTU parameter shall be set to the *maximum size* of the
> > > > > Attribute protocol PDU that the server can receive.
> > > > > The server and client shall set ATT_MTU to the *minimum of the Client Rx MTU
> > > > > and the Server Rx MTU*. The size is the same to ensure that a client can
> > > > > correctly detect the final packet of a long attribute read.'
> > > > >
> > > > > So my interpretation is that the minimum is calculated after the
> > > > > Exchange is complete so it doesn't require the Server MTU to be
> > > > > limited by the Client MTU, anyway if Android and iOS does apply the
> > > > > minimum logic before responding we can do the same.
> > > > >
> > > >
> > > > Maybe it will help if I attach an android packet capture....
> > > >
> > > > The attachedhttps://marc.info/?l=linux-bluetooth&m=164799909020040&w=2
> > > > log can be opened in Wireshark, and to my untrained eye, it looks like
> > > > the MTU negotiation is at packets 451 and 452, and 247 is used?
> > > >
> > > > Ill try and get the proposed patch built and see how it behaves
> > >
> > > Attached if the btmon and bluetoothctl.  It doesnt seem to be there quite yet.
> > >
> > > There was one point before i applied the patch where bluetoothctl
> > > seemed to list the attributes quickly, but i havnt been able to
> > > reproduce that again with or without the patch.
> > >
> >
> > For completeness, also attached the output of my program and btmon
> >
> > My program should connect, wait for resolved signal, then read a char,
> > enable some notifications, and do a write.  You can see the 30 second
> > wait between connect, resolved, and quickly followed by a disconnect.
>
>
> < ACL Data TX: Handle 3585 flags 0x00 dlen 7
>
>                            #22 [hci0] 12.507730
>       ATT: Exchange MTU Request (0x02) len 2
>         Client RX MTU: 517
> > ACL Data RX: Handle 3585 flags 0x02 dlen 7                                                                                                                             #23 [hci0] 12.556081
>       ATT: Exchange MTU Request (0x02) len 2
>         Client RX MTU: 247
> < ACL Data TX: Handle 3585 flags 0x00 dlen 7
>
>                            #24 [hci0] 12.556256
>       ATT: Exchange MTU Response (0x03) len 2
>         Server RX MTU: 247
> > ACL Data RX: Handle 3585 flags 0x02 dlen 7                                                                                                                             #27 [hci0] 12.655972
>       ATT: Exchange MTU Response (0x03) len 2
>         Server RX MTU: 252
>
> So we are now responding with 247 and the peripheral responds with 252
> to our request :/, so we would need to probably delay our request or
> something, that said we are the central so I really which we don't
> have to do it since that would penalize well behaving devices, could
> we perhaps inform the vendor about these issues? I'd considered this
> to be really bad behavior from the remote side since they are not
> following the spec in many respects as even their MTU changes from
> request to response and it does multiple rounds of Exchange MTU and
> stop responding while doing it.

Btw, I checked the android logs and it doesn't even do any Exchange
MTU on its own, so it just responds with 247 so I think this is
related to the peripheral inconsistency with its own MTU (247 != 252)
so it attempts to do another round of Exchange MTU while ATT request
is in place which perhaps cause it to stop responding.

-- 
Luiz Augusto von Dentz