This is a system daemon implementing the TPM2 access broker (TAB) & Resource
Manager (RM) spec from the TCG. The daemon (tpm2-abrmd) is implemented using
Glib and the GObject system. In this documentation and in the code we use
tpm2-abrmd and tabrmd interchangeably.
Instructions to build and install this software are available in the INSTALL.md file.
tpm2-abrmd is a daemon. It should be started as part of the OS boot process.
Communication between the daemon and clients using the TPM is done with a
combination of DBus and Unix pipes. DBus is used for discovery, session
management and the 'cancel', 'setLocality', and 'getPollHandles' API calls
(mostly these aren't yet implemented). Pipes are used to send and receive
TPM commands and responses (respectively) between client and server.
The daemon owns the com.intel.tss2.Tabrmd name on dbus. It can be configured to connect to either the system or the session bus. Configuring name selection would be a handy feature but that's future work.
Check out the man page TPM2-ABRMD(8) for the currently supported options.
This repository also hosts a client library for interacting with this daemon. It is intended for use with the SAPI library (libtss2-sapi) like any other TCTI. The initialization function for this library is hard coded to connect to the tabrmd on the system bus as this is the most common configuration.
Check out the man page TSS2-TCTI-TABRMD(7) and TSS2_TCTI_TABRMD_INIT(3).
The current implementations are mostly equivalent with a few differences. Both provide isolation between objects & sessions created by different connections which is the core functionality required by applications. The reason we have both is that the in-kernel RM was only added very recently (4.12) and we have TPM2 users in environments with kernels going back to the 3.x series. So the user space RM will be around at least till everyone is using the kernel RM.
For the short term we're recommending that developers stick to using the tabrmd as the default to get the most stable / widest possible support. If you structure your code properly you'll be able to switch in / out TCTI modules with relative ease and migrating to the in-kernel RM should be pretty painless. Eventually, all of the required features will end up in the kernel RM and it will become the default.
How we get to the ideal future of a single RM in the kernel: our current plan is to prototype various features in user space as a way to get them tested / validated. There's a lot of stuff in the related TCG spec that we haven't yet implemented and we all agree that it's generally a bad ideal to to put features into the kernel before we:
- understand how they work
- how they're going to be used by applications
- agree we want the feature at all
A good example of this are the asynchronous portions of the SAPI. Right now with the kernel RM you can use the async API but it won't really be asynchronous: Calls to functions that should be async will block since the kernel doesn't supply user space with an async / polling I/O interface. For the short term, if you want to use the SAPI in an event driven I/O framework you will only get async I/O from the user space resource manager. In the long run though, if this feature is important to our users, we can work to upstream support to the in-kernel RM. The plan is to treat future features in the same way.
This was the subject of a talk that was given @ the Linux Plumbers Conference 2017: https://linuxplumbersconf.com/2017/ocw//system/presentations/4818/original/TPM2-kernel-evnet-app_tricca-sakkinen.pdf