Re: [PATCH BlueZ v2] src/gatt-client: always check properties in WriteValue

[David Lechner <david@lechnology.com>]

On 4/28/20 12:03 PM, Luiz Augusto von Dentz wrote:
> Hi David,
> 
> On Tue, Apr 28, 2020 at 7:39 AM David Lechner <david@lechnology.com> wrote:
>>
>> On 4/27/20 11:28 AM, Luiz Augusto von Dentz wrote:
>>> Hi David,
>>>
>>> On Mon, Apr 27, 2020 at 8:07 AM David Lechner <david@lechnology.com> wrote:
>>>>
>>>> This modifies the GATT client characteristic WriteValue D-Bus method to
>>>> always check that the characteristic supports the requested type of
>>>> write by checking for the corresponding property before attempting to
>>>> write.
>>>>
>>>> Before this change, if the "type" option was used and it was set to
>>>> "reliable" or "request", then BlueZ would attempt the write even if the
>>>> characteristic does not support that write type. On the other hand, if
>>>> "type" was set to "command" or was not specified, the method would
>>>> return a org.bluez.Error.NotSupported error without attempting to write.
>>>>
>>>> After this change, the WriteValue method will consistently return
>>>> org.bluez.Error.NotSupported if the corresponding property flag is not
>>>> set for all types of writes.
>>>> ---
>>>>
>>>> v2 changes:
>>>> - remove extra check of test variable not NULL
>>>> - fix one line over 80 chars
>>>>
>>>>    src/gatt-client.c | 8 ++++----
>>>>    1 file changed, 4 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/src/gatt-client.c b/src/gatt-client.c
>>>> index a9bfc2802..f80742fbb 100644
>>>> --- a/src/gatt-client.c
>>>> +++ b/src/gatt-client.c
>>>> @@ -1016,8 +1016,8 @@ static DBusMessage *characteristic_write_value(DBusConnection *conn,
>>>>            *     - If value is larger than MTU - 3: long-write
>>>>            *   * "write-without-response" property set -> write command.
>>>>            */
>>>> -       if ((!type && (chrc->ext_props & BT_GATT_CHRC_EXT_PROP_RELIABLE_WRITE))
>>>> -                       || (type && !strcasecmp(type, "reliable"))) {
>>>> +       if ((!type || !strcasecmp(type, "reliable")) && chrc->ext_props &
>>>> +                               BT_GATT_CHRC_EXT_PROP_RELIABLE_WRITE) {
>>>>                   supported = true;
>>>>                   chrc->write_op = start_long_write(msg, chrc->value_handle, gatt,
>>>>                                                   true, value, value_len, offset,
>>>> @@ -1026,8 +1026,8 @@ static DBusMessage *characteristic_write_value(DBusConnection *conn,
>>>>                           return NULL;
>>>>           }
>>>>
>>>> -       if ((!type && chrc->props & BT_GATT_CHRC_PROP_WRITE) ||
>>>> -                       (type && !strcasecmp(type, "request"))) {
>>>> +       if ((!type || !strcasecmp(type, "request")) && chrc->props &
>>>> +                                               BT_GATT_CHRC_PROP_WRITE) {
>>>>                   uint16_t mtu;
>>>>
>>>>                   supported = true;
>>>> --
>>>> 2.17.1
>>>
>>> As far I can remember this is on purpose so the application can
>>> overwrite the type if it needs to disabling the checks on the client
>>> side, though the kernel can still return not supported, so if we want
>>> to change this it will now not be possible to overwrite it anymore.
>>>
>>
>> If this is the case, does it make sense to make the following change
>> instead so that write without response can also be forced?
>>
>> diff --git a/src/gatt-client.c b/src/gatt-client.c
>> index f80742fbb..3153f571f 100644
>> --- a/src/gatt-client.c
>> +++ b/src/gatt-client.c
>> @@ -1050,8 +1050,8 @@ static DBusMessage *characteristic_write_value(DBusConnection *conn,
>>                           return NULL;
>>           }
>>
>> -       if ((type && strcasecmp(type, "command")) || offset ||
>> -                       !(chrc->props & BT_GATT_CHRC_PROP_WRITE_WITHOUT_RESP))
>> +       if ((type && strcasecmp(type, "command")) || offset || (!type &&
>> +                       !(chrc->props & BT_GATT_CHRC_PROP_WRITE_WITHOUT_RESP)))
>>                   goto fail;
>>
>>           supported = true;
> 
> Yes, that is probably one of the use cases to use the type to force
> sending a command if the client don't care about the response but the
> server don't mark write without response as supported, is that what
> you are after? That sounds like a bug since it appears the intention
> was to allow commands all along.
> 

I just noticed the inconsistency here while trying to implement
writeValueWithResponse() and writeValueWithoutResponse() in the Web
Bluetooth API for the Chromium browser and thought that it should
probably all be one way or the other.

FWIW, in the particular case I am working with, the device has both
the WRITE and WRITE_WITHOUT_RESP properties set so I am using "type"
to force writing without response since write with response would
be selected otherwise.

I will send a proper patch to fix the write without response case.