Home Computer Science A Practical Guide to TPM 2.0
Unlike the IBM TPM 1.2 simulator, the current Microsoft TPM 2.0 simulator, available to TCG members, has no tracing facility. You can't simply run the application and read the simulator's output. It's also unclear whether the TSS implementation will have a tracing capability. Trousers, the TPM 1.2 TSS, had little beyond command and response packet dumps.
However, the simulator source is available. The process is thus the same for nearly any application bug:
1. Run the application to failure, and determine which TPM command failed.
2. Run the simulator in a debugger, and set a breakpoint at the command. Each TPM 2.0 command has a C function call that has exactly the same name as the Part 3 command.
3. Step through the command until the error is detected.
4. It may be necessary to run again, stepping into the Part 4 subroutines, but our experience is that this is often unnecessary.
This section presents some TPM 1.2 application bugs in the hope that they may carry over to 2.0. We also list a few new anticipated error possibilities for TPM 2.0.
TPM 2.0 plain text password authorization should be straightforward. However, HMAC authorization failures are common. The approach is the usual “divide and conquer.” Trace the command (or response) parameter hash, the HMAC key, and the HMAC value. If the hash differs, the parameters that were used in the calculation were different from those sent to the TPM. If the HMAC keys differ, most likely the wrong password was used or the wrong entity was specified. If the HMAC value differs, either it was passed in differently or the salt or bind value was wrong.
Perhaps the most common TPM 1.2 error we've seen is trying to use a disabled function. TPM 1.2 had disabled and deactivated flags, and TPM 2.0 has the corresponding hierarchy enabled.
TPM 2.0 has an additional HMAC error case: the entity may have been created in a way that disallows HMAC authorization. See the attributes userWithAuth and adminWithPolicy in Part 1 of the TPM 2.0 specification.
A typical TPM 1.2 misunderstanding was that creating a key simply returned the key wrapped with the parent—that is, encrypted with the parent's key. It doesn't actually load the key into the TPM; a separate command is required to do that.
TPM 2.0 has an additional case. The TPM 1.2 SRK was inherently persistent.
TPM 2.0 primary keys are transient and must be made persistent. Thus, primary keys may be missing after a reboot.
Finally, an object may have been loaded but is no longer there. You can break at the flush call (or add a printf to the flush call) and the failing command to determine when the object was flushed.
Similarly, a common error for TSS 1.2 was a resource leak—objects (or sessions) were loaded and not flushed, so the TPM eventually filled all its slots. Tracking the load and flush pairs should expose the missing flush, and this will also be true for TSS 2.0.
|< Prev||CONTENTS||Next >|