Support loading kernel, device tree, trust cache from im4p files

[qmfrederik]

This adds support for extracting the kernel, device tree and trust cache from im4p files.

It turns out qemu doesn't link with OpenSSL after all, but does use GnuTLS which uses libtasn1, which supports parsing ASN.1 files as well.

It does require you to define a schema for the ASN.1 structure which you're parsing (hence the img4.asn1 file), which is then converted into a "definitions array" using asn1Parser -o img4.c -n img4_definitions_array img4.asn1.

The code assumes:

• The file is an IM4P file if asn1_der_decoding can successfully parse the file.
• The payload is LZFSE-compressed if it starts with the string bvx.

[qmfrederik]

@TrungNguyen1909 For some reason this doesn't work with the trust cache; so I disabled IM4P support for the trust cache for now.

[TrungNguyen1909]

trust caches use "rtsc" as payload type AFAIK, can you verify if it works or not?

[TrungNguyen1909]

Also, you wouldnt want qemu exits if im4p parser failed because there are cases where you want qemu to load the unpacked data as it.

For instance, ramdisk: its filesystem need to be decompressed. In that case, you wouldnt want to repack it into im4p format, right?

[qmfrederik]

@TrungNguyen1909 , correct, let me rephrase that a bit:

The code assumes:

• The file is an IM4P file if asn1_der_decoding can successfully parse the file. If it is not an IM4P file, the file is loaded as is.
• The payload is LZFSE-compressed if it starts with the string bvx. If the payload does nto start with bvx, the payload is loaded as is.

[qmfrederik]

@TrungNguyen1909 I added support for the trustcache and a couple of sanity checks (to make sure the data which is being loaded is a valid MACH object or a valid trust cache).

I didn't implement the ram disk since it needs to be decompressed anyway.

[TrungNguyen1909]

Errors in extract_im4p_payload should be returned to the caller instead of exit(EXIT_FAILURE).

And when it fails, we should fallback to load the file as it.

[qmfrederik]

Errors in extract_im4p_payload should be returned to the caller instead of exit(EXIT_FAILURE).
And when it fails, we should fallback to load the file as it.

Thanks, in that case, I'd suggest to change the semantics of extract_im4p_payload to be closer to those of g_file_get_contents. That would mean returning TRUE/FALSE depending on whether the operation succeeds.

If the file is not an DER-encoded ASN.1 file which conforms to the IM4P payload specification, the raw content is always returned (as is the case in the current implementation).

But I'm not sure that falling back to the raw payload is always the right thing to do.

For example, when asn1_der_decoding succeeds, the file is a DER-encoded ASN.1 file which conforms to the IM4P payload specification. This means the raw content can never be a valid kernel, trust cache, device tree or ram disk file. Returning the raw content will only result in a failure later down, right?

In that case, I'd suggest to print an error message (for example: "expeting a krnl file but we got a dtre file") and just return FALSE, so that the calling code can clean up and exit.

What about we change the semantics like this:

• Change extract_im4p_payload to return a gboolean just like g_file_get_contents does, and have it return FALSE if g_file_get_contents fails.
• Failures in asn1_array2tree and asn1_create_element are some corner cases, since this code is not operating on user input and not supposed to fail. If they fail, you probably hit an out of memory situation or something similar. So we could probably just assert instead`.
• If asn1_der_decoding fails, the file is returned "as is" (this is already the case)
• If asn1_der_decoding succeeds, but any of the ASN.1-manipulating functions fail, the file is very probably invalid (e.g. the wrong type, wrong magic,...). So call error_report but return FALSE instead of calling exit(EXIT_FAILURE)?

Though in the end, all calls to extract_im4p_payload will probably just abort if extract_im4p_payload returns FALSE (just like they abort now if g_file_get_contents returns false)?

[TrungNguyen1909]

Ah, I got it.

Sorry I missed that earlier.

Thanks.