Don't use PGP/GPG signatures in mail that contains patches

________________________________

On Mon, Jan 18, 2016 at 03:27:15AM +0200, Andrey Utkin wrote:

===== QUOTE =====
Don't use PGP/GPG signatures in mail that contains patches.
This breaks many scripts that read and apply the patches.
(This should be fixable.)
===== END QUOTE =====

This is in Linux' Documentation/email-clients.txt since 2007, and still
almost nobody signs patch submissions. There are few brave people who
do, though, and seems it's not the end of world for any "scripts".
The broken scripts could be an excuse in 2007, but not today.

Proposal:
1. Implement signing option in git-send-email.
2. Figure out if anything fails to interoperate.
3. Drop the quoted statement or change it to appreciate signing.

I don't know about other receiving scripts, but "git am" will handle
signed PGP-MIME out of the box (I didn't try it with inline signatures,
but I imagine it would stick the "BEGIN PGP MESSAGE" cruft into the
commit message).

However, there's an open question of what to _do_ with such a signature.
The email signature does not function as a valid git commit signature.
So you are left with one of:

  1. The receiver can verify the origin of the email before applying the
     patch.

  2. The receiver can keep a copy of the email "somewhere", so people
     can later re-verify it, and then hand-verify that it matches what
     got applied.

     That "somewhere" may just be a mailing list archive, but you could
     get fancy with scripts and associate it with the applied commit
     (e.g., using "git notes").

But those are really questions for the project. If you are mailing your
patches to Linus, does he actually care about (1)? My general impression
of his past opinion is that it's more important to read the patch text
than the "From" line. Of course subsystem maintainers and other projects
may have different opinions.

I think (2) is more compelling, if only to create a better record in the
mailing list archive. Assuming the receivers of your patches don't mind
(and I know some people really _don't_ like things like PGP-MIME,
because their mail readers are not good at replying in-line to the
patches then), I don't it would be a bad thing to teach git-send-email
an option to send it.

-Peff

On 18.01.2016 23:48, Jeff King wrote:

I don't know about other receiving scripts, but "git am" will handle
signed PGP-MIME out of the box (I didn't try it with inline signatures,
but I imagine it would stick the "BEGIN PGP MESSAGE" cruft into the
commit message).

However, there's an open question of what to _do_ with such a signature.
The email signature does not function as a valid git commit signature.
So you are left with one of:

  1. The receiver can verify the origin of the email before applying the
     patch.

  2. The receiver can keep a copy of the email "somewhere", so people
     can later re-verify it, and then hand-verify that it matches what
     got applied.

     That "somewhere" may just be a mailing list archive, but you could
     get fancy with scripts and associate it with the applied commit
     (e.g., using "git notes").

But those are really questions for the project. If you are mailing your
patches to Linus, does he actually care about (1)? My general impression
of his past opinion is that it's more important to read the patch text
than the "From" line. Of course subsystem maintainers and other projects
may have different opinions.

I think (2) is more compelling, if only to create a better record in the
mailing list archive. Assuming the receivers of your patches don't mind
(and I know some people really _don't_ like things like PGP-MIME,
because their mail readers are not good at replying in-line to the
patches then), I don't it would be a bad thing to teach git-send-email
an option to send it.

Thank you for thoughtful reply!
Surely email submission signature cannot be used as git commit
signature. And surely there are issues of usability. And surely people
are used not to care.
But still, if we encourage signing maillist correspondence, we would
avoid impersonation attacks. Imagine that somebody sends stupid
submissions from your name, maintainers shout at you, and your
reputataion is... changed. Of course, you will be able to sort things
out after you read the replies and reply that it's not you. But, given
to openness of maillists, the attacker is able to follow your replies
and insert his ones. Or to reply to your valid submissions that they are
not from you.
Still it seems that making fun of that is not much harder than
masquerading on GitHub
(https://github.com/amoffat/masquerade/commit/9b0562595cc479ac8696110cb0a2d33f8f2b7d29)
Sure there are anti-spoofing measures like DKIM. Honestly I am not aware
if vger.kernel.org is so restrictive that it accepts only letters from
super-safe email servers, but I guess it is not, because not everybody
has this stuff configured on their email servers.

Andrey Utkin [off-list ref] wrote:

But still, if we encourage signing maillist correspondence, we would
avoid impersonation attacks. Imagine that somebody sends stupid
submissions from your name, maintainers shout at you, and your
reputataion is... changed. Of course, you will be able to sort things
out after you read the replies and reply that it's not you. But, given
to openness of maillists, the attacker is able to follow your replies
and insert his ones. Or to reply to your valid submissions that they are
not from you.

Is impersonation an actual problem on vger.kernel.org lists?
Reply-to-all conventions on these lists do a great deal to discourage them.