Re: [PATCH 0/7] Flags and config to sign pushes by default

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Dave Borowitz <dborowitz@xxxxxxxxxx> writes:

> Remembering to pass --signed to git push on every push is extra typing that is
> easy to forget, and just leads to annoyance if the remote has a hook that makes
> signed pushes required. Add a config option push.gpgSign, analogous to
> commit.gpgSign, allowing users to set this flag by default.
>
> Since --signed push will simply fail on any remote that does not advertise a
> push cert nonce, actually setting this to true is not very useful (except for
> the super-paranoid who would never want to push to a server that does not
> support signed pushes). So, add a third state to this boolean, "if-possible",
> to sign the push if and only if supported by the server. To keep parity between
> the config and command line options, add a --signed-if-possible flag to git
> push as well.
>
> The "if-possible" name and weird tri-state boolean is basically a straw man,
> and I am happy to change if someone has a clearer suggestion.

Yes, it looks somewhat strange.  Let me go on a slight tangent to
explain why I think it is OK for "push --signed".

First imagine the case where we were talking an optional setting for
"git tag -a" and what our reaction would be.

Because the reason "git tag -s" would fail when "git tag -a" can
succeed can only be because you do not have a working GPG set-up
(i.e. correctly built and installed GPG with a usable key of your
own to sign), I would say "please sign this tag if I can, but do not
bother failing, an unsigned annotated tag is also OK for me" is not
a sensible request.  In such a case, you'd better get your act
together and make yourself ready to sign before doing the "tag -s"
thing.  Otherwise you'd never get around to do it.  So I'd say "git
tag --sign-if-possible" would not make sense.

But "push --signed" can fail even if you have a perfectly good
GPG set-up.  It will not succeed until the receiving end becomes
ready to accept a signed push, and often you would not be in control
of the receiving end.

More importantly, the meaning and the purose of the GPG signature in
signed tags and signed pushes are vastly different.  In the former
case, you are attesting that the signed objects were made by you to
help yourself.  If somebody else created a tag or a commit and
claimed it is from you, you can say "that signature does not match,
it is not mine".

But "signed push" is not about helping you.  It is about helping the
receiving end by allowing them to be more credible when they say
"This is what David Borowitz said he wanted to put at the tip of
this branch" to other people.  Currently, they can only make a weak
claim "Well, you know, the push was made after we authenticated a
pusher with our own authentication methond, and here is the log that
says the pusher was David".  The log entries could be faked, and the
general public cannot audit.  With a signed push, they can say "Here
is the push certificate, dated and signed by David Borowitz", and
the general public can check without trusting the receiving end
(i.e. hosting site).  If the receiving end does not offer signed
pushes, it just means that they are not ready to be helped by you,
and you should have the option of pushing without helping them,
which is what your "if-possible" is about.

Because of the above reasoning, I think a weaker "I want to do a
signed push if the recipient is capable of accepting one, but
otherwise just pushing there is OK" is a perfectly reasonable
request.

So I am fine as long as "if-possible" turns a failure to make signed
push into a success _only_ when the reason of the failure is because
we did not see the capability supported by the receiving end.  If
the reason why you cannot do a signed push is because you cannot
sign push certificate, "if-possible" should still fail.

Thanks.
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Kernel Development]     [Gcc Help]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [V4L]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Fedora Users]