Re: Patch attestation RFC + proof of concept

[Konstantin Ryabitsev <konstantin@linuxfoundation.org>]

On Wed, Feb 26, 2020 at 05:04:42PM -0400, Jason Gunthorpe wrote:
> > - developer does all their work on a remote VM that doesn't have 
> > access to their PGP keys and submits actual attestation when they 
> > get back to their workstation
> 
> Unfortunately this is a challenging work flow for a lot of reasons. :(

Can you describe why? I would expect that this is done fairly routinely 
due to having more compute power on a remote VM to run various tests.

> > - developer configures their smartcard to require a PIN during each 
> > operation and disables the pgp-agent; sending a series of 40 patches 
> > requires a single PIN entry instead of 40
> 
> This is certainly my situation, my PGP key lives in a yubikey token
> configured for physical presence to confirm. :\

I'm in the same boat. :)

> > - developer submits a v1 of the patch that they don't expect to pass on
> >   the first try and doesn't bother submitting attestation; shockingly,
> >   the maintainer accepts it as-is and the developer can attest their 
> >   patches post-fact *without* needing to collect all the acked-by's 
> >   reviewed-by's etc from all others who have already responded to the v1 
> >   submission
> 
> But there won't be tags in this case, so how do we learn the original
> attestation-id to find the detatched signature?

The attestation would be performed before all the follow-up tags are 
applied, so the attestation-id would be the same. After the patch is 
applied to a git repository, we can still use the "i" hash to look it up 
(see more below).

> > For example, a maintainer will almost certainly edit the message 
> > content to add their own Signed-off-by, and may edit the patch for 
> > minor nitpicking. 
> 
> The i/p/m will always change once in git:
>    - The commit message is always changed, at least to add sign off
>    - The email Subject is always changed to strip [PATCH xxx]

This is already done by "git mailinfo" so I would expect that the i: 
hash almost never changes, actually, unless the maintainer actually 
edits the subject. Subject + Author + Email are sufficiently unique to 
be able to locate the attestation data of the exact patch.

> If we know the expected ID then you could do some fuzzy scheme to
> cancel out, or at least check, the differences..
> 
> We have many patches now that have Link: trailer. I think it would be
> useful to run some analysis:
>  - Fetch the original patch email from the Link and compute the
>    detached signature hashes
>  - Attempt to validate this sigature against the git commit
>  - Is there a fuzzy algorithm that brings this rate higher?

So, the goal is not really to perform attestation once the patches made 
it into the git tree. I am specifically trying to address the following 
cases:

- Someone poses as a trusted developer and submits a malicious patch
- Someone bribes me to edit a patch on lore.kernel.org to introduce a 
  backdoor

Currently, maintainers have no mechanism to check against either of 
these cases. Adopting end-to-end patch attestation resolves both 
problems.

> > Full i-m-p attestation will fail in this case, but we can then query 
> > the signatures archive for each individual hash to identify which 
> > part of the submission fails validation:
> > https://lore.kernel.org/signatures/?q=2a02abe02216f626105622aee2f26ab10c155b6442e23441d90fc5fe4071b86e
> >
> > This lets us present the maintainer with more useful info, like: "full 
> > attestation failed, but the only changed part is actually the message 
> > and not the patch content, so it's probably still okay to apply."
> 
> How does the message body get changed in transit from the submitter to
> the maintainer?

It shouldn't, because this means that the patch becomes mangled. We do 
perform a single normalization routine, which is converting all '\r\n' 
into '\n'. Any other changes done to bodies mean broken patches.

-K