Re: Bare repositories in the working tree are a security risk

[Glen Choo <chooglen@google.com>]

Thanks a lot for chiming in, especially for advocating on behalf of
users of embedded bare repos :)

"brian m. carlson" <sandals@crustytoothpaste.net> writes:

> On 2022-04-07 at 21:53:26, Justin Steven wrote:
>> Hi all,
>> 
>> I'm the author of one of the articles linked in Glen's mail. Thank you
>> Glen for summarising the problem beautifully and pushing this forward.
>> 
>> Brian said:
>> > As mentioned elsewhere, git status doesn't work without a working tree.
>> 
>> This is correct. However, it is possible to embed a bare repo that has
>> its own core.worktree which points to a directory within the
>> containing repo, satisfying the requirement of having a working tree.
>> This is covered in the article [1] and looks to be accounted for in
>> Taylor's reproducer script which admittedly I haven't run.
>> 
>> > Instead, I'd rather see us avoid executing any program from the config
>> > or any hooks in a bare repository without a working tree (except for
>> > pushes).  I think that would avoid breaking things while still improving
>> > security.
>> 
>> Due to the fact that the embedded bare repo can be made to have a
>> working tree, this won't be an effective fix.
>
> Then we'd probably be better off just walking up the entire hierarchy
> and excluding worktrees from embedded bare repositories, or otherwise
> restricting the config we read.  That will probably mean we'll need to
> walk the entire directory hierarchy to see if it's embedded (or at least
> to the root of the device) in such a case, but that should be relatively
> uncommon.

IIUC, the difference between your suggestion and (3) in the original
email:

  == 3. Detect that we are in an embedded bare repo and ignore the embedded bare
  repository in favor of the containing repo.

is that we only walk the hierarchy in the event that the bare repo has a
worktree? If so, then yes that seems relatively uncommon and wouldn't
break most bare repo users.

What I'm unsure about is whether or not this reduces the attack surface
enough; it protects the highest value target, `git status`, but I'm not
sure about the following:

* How well does this protect other, existing commands (e.g. `log`,
  `branch` and `remote` fall into the same category of commands that
  appear informative and harmless)?
* How well does this protect other, abusable configuration directives
  (`core.pager` perhaps)?
* How future-proof is this? (if someone removes a dependency on
  core.worktree, is the command vulnerable again?)

There's a similar argument to be made about restricting the config we
read - we'd be running the risk of either missing some abusable config
that already exists or will exist.

And if my understanding is correct (and it very well might be wrong),
then this seems too piecemeal to be worth the effort.

> I'd definitely like to see us make a security improvement here, but I
> also would like to avoid us breaking a lot of repositories, especially
> since we lack alternatives.

Agreed, leaving existing users high and dry is an awful outcome.

>
> If git fast-import could 100% correctly round-trip all commits and
> repositories, I would be much more open to blocking this in fsck after a
> deprecation period, but as it stands that's not possible.  Perhaps
> improving that would be a suitable way forward.