injected body trailers

[Kees Cook]

On Thu, Oct 21, 2021 at 04:44:59PM -0400, Konstantin Ryabitsev wrote:
> On Thu, Oct 21, 2021 at 01:22:31PM -0700, Kees Cook wrote:
> > Hi!
> > 
> > So, I just saw a DKIM failure, and it was entirely justified. :)
> > 
> > Grabbing thread from lore.kernel.org/all/20211021142516.1843042-1-ardb%40kernel.org/t.mbox.gz
> > Checking for newer revisions on https://lore.kernel.org/all/
> > Analyzing 1 messages in the thread
> > Checking attestation on all messages, may take a moment...
> > ---
> >   ✓ [PATCH] ARM: stackprotector: prefer compiler for TLS based per-task protector
> >     ✓ Signed: openpgp/ardb at kernel.org
> 
> You will notice that the openpgp signature passed. This is because we:
> 
> 1. record the length of the original message when we're creating the signature
>    (see l=2495 in X-Developer-Signature)
> 2. if the initial validation fails and the body is longer than l=2495, we trim
>    the body to that number of bytes
> 3. if the trimmed validation passes, we use that version for the patch body
>    content, since that's clearly what the developer intended

I suspected something like this was happening to make that one pass.
Nice.

> 
> >     ✗ BADSIG: DKIM/kernel.org      
> >     ✓ Signed: DKIM/lists.infradead.org (From: ardb at kernel.org)
> > ---
> > 
> > This is https://lore.kernel.org/all/20211021142516.1843042-1-ardb@kernel.org/
> > and for some reason, the linux-arm-kernel mailing list is injecting a
> > body trailer.
> 
> "For some reason" is really "that's the default for mailman-2". Mailman-2
> belongs to a wholly different era and *can* be configured to be DKIM
> compliant, but rarely is.
> 
> > I just downloaded this directly and removed the trailer, and the DKIM
> > passed. This experience has raise a few questions...
> > 
> > 1) Can (should) b4 grow logic to progressively strip lines off the end
> >    of a body until DKIM passes?
> 
> Ah, but then the lists.infradead.org DKIM will fail. Theoretically, we should
> always prioritize the signature that is closest aligned with the From: header,
> but that's not actually that straightforward, as DNS lookup and validation
> rules can get really complex.

Could each signature validation independently process the body, with
the smallest signed body being what is "produced"? i.e. GPG already
self-trims. DKIM could do the same, trying to find a matching body i.e. on
failure (slow path), trying trimming up to 10(?) lines progressively
looking for a match?

(Probably better is to just fix the mailing lists, but maybe this would
be useful for historical patch extraction? Dunno.)

> 
> > 2) Can the linux-arm-kernel mailing list please stop breaking DKIM?
> >    Who should authorize this change (rmk, Catalin)? And who can make
> >    the change (peterz)?
> 
> The relevant settings should be a) don't add any subject prefixes, b) don't
> add anything to the body trailers, c) don't rewrite any other headers (to, cc,
> reply-to, etc).

rmk, Catalin, Peter, can this get sorted out? Having mailing list
trailers is annoying beyond just DKIM breakage. :)

> 
> >    (I realize now that all the mail from linux-arm-kernel has been
> >    getting dropped into my Spam folder -- I normally don't notice since
> >    I'm usually CCed directly or via some other list on things I wanted
> >    to see.)
> > 
> > 3) Are there other lists for which lore is collecting emails where DKIM
> >    is persistently broken, and can we fix those lists too?
> 
> I would also note that lists.infradead.org should not really be adding its own
> DKIM signature to messages it sends out. It doesn't really serve any purpose
> unless the From: header is rewritten (but please don't do that either).

-Kees

-- 
Kees Cook