DBus error handling in BlueZ is a mess. This is the first patch to unify
all DBus error handling like in ConnMan and oFono. This unifies all
.InvalidArguments errors.
If the GIOChannel is in the buffered state (the default) the watch
function is called without receiving a POLLOUT from the socket. GLib
adds a G_IO_OUT condition just because there is space in the GIOChannel
internal buffer.
The solution is disabling the internal buffer, which in turn, makes the
call of g_io_channel_flush() useless.
Attributes may require encryption for certain operations. This commit
adds checks to the attribute server which verify whether the current
connection is encrypted (by checking the security level) and the
attribute being accessed requires authentication. If encryption
requirements are not satisfied, the "Insufficient Encryption" error is
returned by the server.
Note that currently there is no actual distinction between
authentication and encryption. Currently, if an attribute requires
authentication, it is only checked whether the link is encrypted. Core
Specification, as of version 4.0, does not specify an explicit
"encryption" permission (see Vol. 3, Part F, 3.2.5 "Attribute
Permissions").
The attribute server must verify if the operation (read/write) is
permitted before running the request, and send "Read Not Permitted" or
"Write Not Permitted" error response to client if appropriate.
Add enums for attribute read/write requirements, which may be "none",
"authentication required" or "authorization required". Additionally, a
"not permitted" requirement means that operation is not permitted,
allowing read-only or write-only attributes.
The attrib_db_add() API was changed to allow setting these requirements,
and the example server was changed to set requirements for its
attributes.
In case service records have been added to bluetoothd before a new
adapter is registered, the records which are shared by all adapters
(indicated by having the address set to BDADDR_ANY) need to be added
to the services list of the new adapter. This patch adds a function
for this on adapter initialization.
The issue could be reproduced by running bluetoothd and obexd on a
PC and briefly removing the BT dongle. The service records from
obexd would not be present in the adapter's local list (which is
used to set the class of device).
Discover All Primary Services should not be trigged by the attribute
client when a given device is registered. Discover services is now done
by the device entity.
If Discover All Primary Services or SDP search fails, the CreateDevice
caller will not receive a response. Error reproducible when a Discover
All Primary Services is sent over a LE link and the kernel doesn't
address properly the connections in the channel ID 4.
For CreateDevice, if the remote device is LE capable, Discover All
Primary Services shall be issued instead of SDP queries. The logic
to define if the remote is LE capable shall be based on the result
of the interleaved discovery results. Meaning, the remote device
is classified as LE capable only if an advertising event from this
device was collected during the discovery.
Limitation: Advertising event data is not being considered yet to
infer the remote properties and CreateDevice needs to be called
during the discovery session.
Discover primary services implemented inside the device entity to allow
proper integration of attribute plugin. Implements a single entry point
to the attribute plugin no matter the transport(BR/EDR or LE), the device
probe callback is called for both types.
Add a new function to discover all primary services without additional
calls to fetch the remaining primary services, sub-procedure iterations
is handled inside this function.
The next action are: clean the attribute client removing implicity service
and characteristics discovery, issue the Discover Primary Service based on
the remote properties and fetch the characteristic on demand.
In addition to blocking the mainloop hci_send_req changes the socket
filters which makes it unusable for the per-adapter HCI socket in
hciops.c. Using hci_send_cmd makes much more sense here.
Previously simultaneous authentication requests to the same device caused
bluetoothd crash. Now if ongoing authentication occurs error is returned,
preventing from simultaneous requests to the same device.
This changes hook telephony driver de/initialization to powered state so
telephony drivers can now free their resources when there is no adapter
active in the system.
This patch lets set_speed function changing UART baud rate to return an
error code if any one operation fails instead of returning only the last
operation's status.
When a io_watcher is added to an MDL or an MCL channel, its reference
should be incremented because the watcher should keep its own reference
the the structure.
Also a destroy function is added in order to decrement the reference
once the watcher is removed.
The value 0x03 isn't something that exists in the core spec, so it
shouldn't be used in the same API as SCO/ACL link types. Since there are
some experimental kernel patches that use it this patch makes it private
to hcitool.c.
Due to a possible kernel bug, sometimes the very first HCI commands that
the kernel sends might go unnoticed. Therefore check for this when one
of the last commands (read_local_name) that the kernel sends completes.
