Dump attested TLS credentials upon starting up #108
Conversation
|
For each failure mode, could you add a SGXLKL_VERBOSE line? |
|
I wonder if this should be an optional feature enabled in the app config since it adds a strict dependency to DCAP and hence network. |
|
[ Duplicating our private discussion here ] We need to address the layering here before it can be merged. We should not be introducing new code into
|
There was a problem hiding this comment.
[ Duplicating our private discussion here ]
We need to address the layering here before it can be merged. We should not be introducing new code into
src/enclavethat has nontrivial dependencies on libc. Probably the best structure for this is to:
- Provide functionality in src/enclave for getting the raw attestation information, without mbedTLS dependencies. This can talk to OE APIs directly.
- Modify LKL to allow exposing that somewhere in sysfs (or possibly as a device node) by calling the APIs added in src/enclave.
@davidchisnall , which flow triggers sysfs here?- Add the userspace component that reads the raw data from the kernel-owned file, uses mbedTLS to generate the certificates, and puts them in a well-known location.
Is it still triggered by startmain in enclave_init.c as in the current change?
| if (private_key_size_out) | ||
| *private_key_size_out = 0; | ||
|
|
||
| if (!cert_out || !cert_size_out || !private_key_out || |
There was a problem hiding this comment.
Not clear about the logics here.
From lines 163-173, all the values pointed by these out parameter pointers are assigned NULL or 0 before the 'if' logic checks pointers and sends the flow to 'done'. Who will be manage the memory pointed by these pointers before nullifying them?
| } | ||
|
|
||
|
|
||
| static int _write_file(const char* path, const void* data, size_t size) |
There was a problem hiding this comment.
Both this (_write_file) and _load_file are pretty general utility functions. Do we have a more common place for them than gencreds.c?
No description provided.