[Opensc-devel] Adding libp11 (or its equivalent) to OpenSSL

[David W. <dw...@in...>]

It would be really useful if OpenSSL *included* support for PKCS#11 as
a first class citizen.

This would mean that it could be natively incorporated into higher
level APIs such as SSL_CTX_use_certificate() and friends. Basically any
API that can take a filename to reference a certificate, should also be
able take a RFC7512 PKCS#11 URI.

This would also allow us to use a coherent trust database from PKCS#11,
which solves the problem of which *purposes* we trust each CA for,
unlike the existing flat-file solutions.

And applications would no longer need to jump through additional hoops
and have additional dependencies to get PKCS#11 support; we could make
it largely Just Work™, like it does for example with GnuTLS.



The code in libp11 is basically written to be OpenSSL code. If you
dropped it into the crypto/pkcs11 directory of OpenSSL precisely as it
stands, it wouldn't look out of place.

I propose — as the starting point of a plan which will surely be
modified by the time we conclude this thread — that we do so.

The biggest barrier to this, of course, is the licence. For reasons
which are lost in the mists of time, libp11 is licensed under the
LGPLv2, and is not compatible with the OpenSSL licence.

Therefore, I propose that we relicense the libp11 project under a
standard 3-clause BSD licence.

I did a bit of a sanity check offline, and I contacted Olaf Kirch as
well as the top contributors responsible¹ for 8611 of the 8984 lines of
.[ch] files in the repository. All of whom so far have agreed to allow
their contributions to be re-used under a BSD licence.

That gives me reasonable confidence that we can all agree, and that if
there *are* some people we can't find, or who don't agree, we can
reimplement the corresponding lines of code without too much of a
problem.

If you have ever contributed to libp11 (or the engine, although I
wasn't going there yet), I would be very grateful for an explicit
response to the question: "May I re-use your code under the 3-clause
BSD licence as seen at https://opensource.org/licenses/BSD-3-Clause?"

(There is a separate question, if we get that far, which is "shall we
*change* the licence of the libp11 project to 3-clause BSD?".
Personally I'd say yes to that too.)



So next we come to the technical approach, and I'd really appreciate
feedback here.

Do we *want* to drop the libp11 API directly into OpenSSL? That might
be ideal from the point of view of existing applications as they
migrate from OpenSSL <=1.1 + libp11, to OpenSSL 1.2.

Or perhaps we want to take this opportunity to change the API? Perhaps
libp11 would continue to exist alongside OpenSSL 1.2 for compatibility
with those existing apps?

(That question is specifically aimed as the OpenSSL folks on Cc as well
as the libp11 list.)


For a start, I definitely think we should at least add functions for
obtaining an EVP_PKEY or an X509 cert from a PKCS#11 URI alone —
similar to the pkcs11_load_key() and pkcs11_load_cert() functions in
the engine code.

I'm also tempted to suggest that we should make it capable of using
p11-kit for the basic module loading and initialisation. Since p11-kit
is "sufficiently ubiquitous" on the platforms where this is relevant,
my approach would probably be to *start* by depending on p11-kit, and
if anyone objects they can do so in 'diff -up' form. Starting with a
full implementation of RFC7512 URI parsing...


Once we have a consensus on what the resulting API would look like, 
my idea would be to create a sequence of submissions to OpenSSL (not
for 1.1 but to be merged to OpenSSL HEAD after that), which slowly add
the required functionality step by step (probably leaving out the
deprecated functions).

Each commit would then be easily reviewable for merging into OpenSSL.
And where it's lifted from existing libp11 code, I could properly check
the provenance of each line of code as I go, to make sure I do have
permission to relicense it.

That's about as far as my plan goes, really. The point here is to
solicit feedback and work out how best to proceed...

Please note there are two members of the OpenSSL team in Cc. Please
make sure you "reply to all" and don't drop them.

-- 
dwmw2


¹ According to 'git blame', which I appreciate is a blunt tool and doesn't
  even credit Olaf with any lines of code. But as a sanity check for "should
  I give up on this idea and just write something from scratch?" it's OK.
  Specifically:

$ for a in `find . -name \*.[ch]` ; do git blame $a; done | sed 's/........ (\(.*\)[[:space:]]*20..-..-.. ..:..:.. .*/\1/' | sed s/[[:space:]]*$// | sort | uniq -c  | sort -rn
   4559 Michał Trojnara
   2813 Andreas Jellinghaus
    749 Doug Engert
    454 Nikos Mavrogiannopoulos
    163 Ludovic Rousseau
     59 Mikhail Denisenko
     45 Alon Bar-Lev
     44 Stephane Adenot
     39 Nils Larsch
     32 Michal Trojnara
      9 Martin Paljak
      6 Anton Fedorov
      4 David Woodhouse
      3 Christian Heimes
      2 Mike Gerow
      2 Jean-Pierre Szikora
      1 LE TOUX Vincent