Drop checksum field from signed files db

[WhyNotHugo]

This is required before moving files from /usr into their proper
locations.

See: #57

[WhyNotHugo]

Still in draft due to:

• One big comment in the middle regarding which feedback is welcome.
• One test is still failing.
• Tests need a locally EFI binary as input. I'm using a system one, but then tests will fail in other setups.

[WhyNotHugo]

Regarding this bit

sbctl/keys.go

Lines 150 to 188 in e742a7f

	func SignFile(key, cert, file, output string) error {
	// Check file exists before we do anything
	if _, err := os.Stat(file); errors.Is(err, os.ErrNotExist) {
	return fmt.Errorf("%s does not exist", file)
	}
	// Let's check if we have signed it already AND the original file hasn't changed
	ok, err := VerifyFile(cert, output)
	if errors.Is(err, os.ErrNotExist) && (file != output) {
	// if the file does not exist and file is not the same as output
	// then we just catch the error and continue. This is expected
	// behaviour
	} else if err != nil {
	return err
	}
	peExisting, err := os.ReadFile(file)
	if err != nil {
	return err
	}
	ctxExisting := pecoff.PECOFFChecksum(peExisting)
	peFile, err := os.ReadFile(file)
	if err != nil {
	return err
	}
	ctx := pecoff.PECOFFChecksum(peFile)
	// XXX: I'm not 100% sure this is the best approach. A file corrupt in
	// the right places might not be regenerated. Maybe it's best to strip
	// the signature part of the destination and compare the non-signature
	// part of the target file with the current one?
	// TBH, this comparison is expensive enough that it might just be worth
	// re-generating the target file even if it's up to date anyway.
	if ok && bytes.Equal(ctxExisting.SigData.Bytes(), ctx.SigData.Bytes()) {
	return ErrAlreadySigned
	}

, I've doubts.

Is stripping the signature and re-signing a valid way to determine "it's already signed"? If the signature mismatches, we can either bail or overwrite. I think exiting non-zero is best?

[Foxboron]

Is stripping the signature and re-signing a valid way to determine "it's already signed"? If the signature mismatches, we can either bail or overwrite. I think exiting non-zero is best?

Can't we use pkcs7.VerifySignature? If we can't verify the file then we probably havent signed it. It seems simpler than messing with the implementation details of the weird context struct :)

[Foxboron]

I realized I might not even correctly remember the contraints we are working with :/ I really need to get back into hacking on all of this.

[WhyNotHugo]

We also need to verify whether the file has changed or not. The file might have a valid signature, but be a different file (e.g.: the previous version of the bootloader).