The current server side channel handling of AUDIO_INPUT is currently
very constrained:
- Server implementations cannot measure the clients uplink, since the
Incoming Data PDU is currently unhandled and FreeRDPs DSP handling
delays the callback call of ReceiveSamples
- Servers currently cannot prefer a different protocol version
- Servers currently cannot change the used format
To solve these issues without running into the risk that some
simplifications constraint certain API usage, rework the current channel
handling to be very close to the documentation.
This means, that all documented API calls can be made by server
implementations and all documented PDUs, that the server side is
expected to receive are just parsed inside FreeRDP and then forwarded to
the API implementation.
The shadow server tries to resize the client to use a compatible
resolution. If that fails, e.g. if there is another resize request after
the disconnect/reconnect sequence, abort the connection.
* Fixed GetFileInformationByHandle initializers
* Fix#7793: Do not expose internal input API
Slow-Path input uses UINT16 for scancodes on wire, but only the
lower byte is actually used. (the extended fields are sent in
keyboardFlags field)
Hide this implementation detail and adjust the API to use UINT8
for the code instead just like the corresponding Fast-Path PDU
* Added a warning for problematic slow path keyCodes
It will be completely possible to implement subsystem with only include/freerdp/server/shadow.h and libfreerdp-shadow.
Details as following:
1. Exported surface structure as subsystem implementations deeply depend on it to send image update
2. Export capture APIs. They are actually indepent APIs to help compare and calculate image difference.
3. Introduce API to trigger client frame update. Conceal details in subsystem->updateEvent
4. Pass client to client callbacks. Subsystem implementation may need to know 'which client' send the interaction event as well as the authentication request.
Add this support in callback definition before anyone really use shadow framework APIs to implement a custom subsystem.
Also added callback for client capability exchange
5. Remove X11_ShadowSubsystem Mac_ShadowSubsystem Win_ShadowSubsystem from libfreerdp-shadow.
Discard FREERDP_API mark on ShadowSubsystemEntry functions and make them be compiled together with shadow.c in CMakeLists.txt.
This is required from PR #2751.
Now subsystem implementations and shadow.c could be regarded as an example for shadow framework.
Detail fixes:
1. Add resize support in shadow framework layer
2. Enhance X11 implementation to detect desktop resolution change
3. Fix the subRect feature.
It seems not completely finished and it looks incompatible in different source code.
Fix it to be consistent in all source code as following:
a. The subRect is only awared in framework layer, subsystem implementation should not be aware of it. It only take effect at shadow_client and corresponding shadow_input.
b. The screen and surface should only represent a monitor screen. They don't need to consider subRect feature.
4. A lobby should be alternative image shown to client when the client is not 'mayView'.
We don't need to have seperate lobby for each client, move it into server structure as 'another surface'
5. Fix display handle leak in enum monitor of X11
6. Suppress BadMatch Error while capturing the image
Original pointer alpha update message is too specific for x11 implementation.
It doesn't allow pointer alpha mask with both 'xor' color and 'and' mask, e.g.: the 'edit' pointer in windows text box.
1. Move shadow_client_convert_alpha_pointer_data to x11 implementation as it is specific for x11.
2. Update message definition to be more generic: to accept 'xor/and' mask directly.
Implementation of subsystem can implement its own way to convert pointer mask data.
3. Fixed fault handling to free the resource allocated.
1. Export fps related API so that subsystem implementation no longer need to know about details in encoder structure.
2. Discard frameList dictionary.
The 'value' in this dictionary is never used and not properly free'ed when client is disconnected.
The dictionary was used to calculate 'inflight' frame count. Once an ACK is received from client, an item in the dictionary is removed.
We then calculate 'inflight' frame by the count of the items in the dictionary.
However, some rdp clients (win7 mstsc) skips frame ACK if it is inactive, ACK of some frame would actually never arrive.
We actually don't need the dictionary. We only need to record the latest acknowledged frame id, and the difference between last sent frame id is the inflight frame count.
3. Minor fix in default fps calculation. encoder->frameAck is wrongly used as integer while it's actually bool flag.
1. Introduce message queue in shadow client. No longer use subsytem->MsgPipe->out to deliver message to clients.
We used to use subsytem->MsgPipe->out for messages which need to be sent to client. But it's not correct. Only one client would get the message if multiple client exists
This problem make the fix in PR #2643 incomplete.
Introduced reference count based solution to release resource taken by the message.
Also added APIs for client message delivery.
Also fixed msg pipe in subsystem to clean resource when destroyed.
2. Discard unused StopEvent in client. We actually use quit message instead.
3. Enhance disposal of channels.
Free context for remdesk and encomsp channels. The original fix only stop the threads, but doesn't release resource.
Dispose channels earlier. The channels are built on client->vcm. Disposing channels after client->vcm is closed cause unknown behavior.
Original fix is #2644
4. Start to add audio support.
