commit | c3632a778f56d99544e598f03b877e8217f8fa7d | [log] [tgz] |
---|---|---|
author | Harald Welte <laforge@gnumonks.org> | Sun Oct 14 20:38:34 2018 +0200 |
committer | Harald Welte <laforge@gnumonks.org> | Mon Oct 15 00:27:20 2018 +0000 |
tree | 4ac9a886b72d1c13c416d416a5a9dc0e24799db4 | |
parent | 098ef8780cb5ec570d86e476f72c490115d83341 [diff] |
rspro_gen_ConnectClientReq(): don't dereference optional 'client' Change-Id: Id5ff7f1524bc34576095d6812935557c7f03b545
This software suite is a work in progress.
The client interfaces with GSM phones / modems via dedicated "Card Emulation" devices such as the Osmocom SIMtrace2 or sysmocom sysmoQMOD board + firmware. This hardware implements the ISO7816-3 electrical interface and protocol handling and passes any TPDU headers received from the phone/modem to remsim-client for further processing of the TPDUs associated to the given APDU transfer.
remsim-client connects via a RSPRO control connection to remsim-server at startup and registers itself. It will receive configuration data such as the remsim-bankd IP+Port and the ClientId from remsim-server.
After receiving the configuration, remsim-client will establish a RSPRO data connection to the remsim-bankd IP:Port.
As the USB interface for remote SIM in simtrace2.git uses one interface per slot, we can implement the client in blocking mode, i.e. use blocking I/O on the TCP/RSPRO side. This simplifies the code compared to a more complex async implementation.
The remsim-bankd (SIM Bank Daemon) manages one given SIM bank. The initial implementation supports a PC/SC driver to expose any PC/SC compatible card readers as SIM bank.
remsim-bankd initially connects via a RSPRO control connection to remsim-server at startup, and will in turn receive a set of initial [client,slot]:[bankd,slot] mappings. These mappings determine which slot on the client (corresponding to a modem) is mapped to which slot on the SIM bank. Mappings can be updated by remsim-server at any given point in time.
remsim-bankd implements a RSPRO server, where it listens to connections from remsim-clients.
As PC/SC only offers a blocking API, there is one thread per PC/SC slot. This thread will perform blocking I/O on the socket towards the client, and blocking API calls on PC/SC.
In terms of thread handling, we do:
The worker threads initially don't have any mapping to a specific reader, and that mapping is only established at a later point after the client has identified itself. The advantage is that the entire bankd can live without any non-blocking I/O.
The main thread handles the connection to remsim-server, where it can also use non-blocking I/O. However, re-connection would be required, to avoid stalling all banks/cards in the event of a connection loss to the server.
worker threads have the following states:
Once the client disconnects, or any other error occurs (such as card I/O errors), the worker thread either returns to INIT state (closing client socket and reader), or it terminates. Termination would mean that the main thread would have to do non-blocking join to detect client termination and then re-spawn clients, so the "return to INIT state" approach seems to make more sense.
Open topics: