Linux has a netlink-based user-space interface for Kernel cryptography. Kernel based crypto has a couple of advantages that are explained at http://www.chronox.de/libkcapi/html/ch01s02.html . The document doesn't mention that a crypto socket also supports splicing and sendfile. Files no longer have to be copied to user-space.

My experimental branch https://github.com/tiran/cpython/commits/feature/af_alg implements af_alg support. Example:

from socket import socket, AF_ALG, SOCK_SEQPACKET, SOL_ALG, ALG_SET_KEY
from binascii import hexlify
with socket(AF_ALG, SOCK_SEQPACKET, 0) as alg:
    alg.bind(('hash', 'hmac(sha512)'))
    alg.setsockopt(SOL_ALG, ALG_SET_KEY, b'key')
    op, _ = alg.accept()
    with open('/etc/passwd', 'rb') as f:
        op.sendfile(f)
    print(hexlify(op.recv(64)))
    op.close()

Working patch with tests and documentation.

socket.algset() isn't strictly necessary but makes the feature much more pleasant to use. I accept ideas for a better name, though.

I reviewed AF_ALG-kernel-crypto-support-for-socket-module.patch.

Thanks for your review, Victor. I have addressed most of your remarks.

* algset() is now called sendmsg_afalg(). It behaves more like a specialized version of sendmsg() and can optionally handle an array of iovec.

* I had to add another variant of setsockopt that sends NULL, int instead of (char*)int, sizeof(int) to get the AEAD GCM tests working. AEAD expects ALG_SET_AEAD_AUTHSIZE as (NULL, taglen). algo.setsockopt(SOL_ALG, ALG_SET_AEAD_AUTHSIZE, (None, taglen)) sends optval=NULL, optlen=taglen.

* Added tests for AES-CBC decryption, AEAD AES-GCM and DRBG.

New patch with setsockopt(socket.SOL_ALG, socket.ALG_SET_AEAD_AUTHSIZE, None, taglen) instead of (None, taglen).

I have removed binascii.(un)hexlify().

New changeset 74ce062a0397 by Christian Heimes in branch 'default':
Issue #27744: Add AF_ALG (Linux Kernel crypto) to socket module.
https://hg.python.org/cpython/rev/74ce062a0397

New changeset 52404f9596b5 by Christian Heimes in branch 'default':
Issue #27744: correct comment and markup
https://hg.python.org/cpython/rev/52404f9596b5

New changeset ee32af890e27 by Christian Heimes in branch 'default':
Issue 27744: Check for AF_ALG support in Kernel
https://hg.python.org/cpython/rev/ee32af890e27

New changeset 4ebe3ade6922 by Christian Heimes in branch 'default':
Issue 27744: AES-CBC and DRBG need Kernel 3.19+
https://hg.python.org/cpython/rev/4ebe3ade6922

Despite the last changes, test_aes_cbc() hangs for fifteen minutes:
http://buildbot.python.org/all/builders/x86-64%20Ubuntu%2015.10%20Skylake%20CPU%203.x/builds/1298/steps/test/logs/stdio
running: test_socket (900 sec)
0:28:47 [332/402] test_socket crashed
Timeout (0:15:00)!
Thread 0x00007f2843d94700 (most recent call first):
  File "/home/buildbot/buildarea/3.x.intel-ubuntu-skylake/build/Lib/test/test_socket.py", line 5389 in test_aes_cbc
  File "/home/buildbot/buildarea/3.x.intel-ubuntu-skylake/build/Lib/test/support/__init__.py", line 523 in wrapper
  File "/home/buildbot/buildarea/3.x.intel-ubuntu-skylake/build/Lib/unittest/case.py", line 600 in run
  . . .

Also, the Gentoo buildbots fail:
http://buildbot.python.org/all/builders/x86%20Gentoo%20Non-Debug%20with%20X%203.x/builds/1368/steps/test/logs/stdio
======================================================================
ERROR: test_aead_aes_gcm (test.test_socket.LinuxKernelCryptoAPI)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/buildbot/buildarea/3.x.ware-gentoo-x86.nondebug/build/Lib/test/support/__init__.py", line 523, in wrapper
    return func(*args, **kw)
  File "/buildbot/buildarea/3.x.ware-gentoo-x86.nondebug/build/Lib/test/test_socket.py", line 5428, in test_aead_aes_gcm
    with self.create_alg('aead', 'gcm(aes)') as algo:
  File "/buildbot/buildarea/3.x.ware-gentoo-x86.nondebug/build/Lib/test/test_socket.py", line 5346, in create_alg
    sock.bind((typ, name))
FileNotFoundError: [Errno 2] No such file or directory

Similar failures for test_aes_cbc test_drbg_pr_sha256 test_hmac_sha1 test_sha256.

> Also, the Gentoo buildbots fail:
> http://buildbot.python.org/all/builders/x86%20Gentoo%20Non-Debug%20with%20X%203.x/builds/1368/steps/test/logs/stdio
> ======================================================================
> ERROR: test_aead_aes_gcm (test.test_socket.LinuxKernelCryptoAPI)

It's Linux 4.4.6. configure says "checking for sockaddr_alg... yes".

New changeset e3b83bfa02c5 by Christian Heimes in branch 'default':
Issue 27744: skip test if AF_ALG socket bind fails
https://hg.python.org/cpython/rev/e3b83bfa02c5

Some distributions mess with the Kernel or disable user-space crypto. I have added some tweaks and fixed a couple of buildbots. I don't know what is going on with x86-64 Ubuntu 15.10 Skylake CPU. It's a Kernel 4.2 machine and should support AES-CBC.

My PC is Ubuntu15.10, kernel 4.2, though CPU not Skylake. Everything works fine.

test_aead_aes_gcm (test.test_socket.LinuxKernelCryptoAPI) ... skipped "('[Errno 2] No such file or directory', 'aead', 'gcm(aes)')"
test_aes_cbc (test.test_socket.LinuxKernelCryptoAPI) ... ok
test_drbg_pr_sha256 (test.test_socket.LinuxKernelCryptoAPI) ... ok
test_hmac_sha1 (test.test_socket.LinuxKernelCryptoAPI) ... ok
test_sendmsg_afalg_args (test.test_socket.LinuxKernelCryptoAPI) ... ok
test_sha256 (test.test_socket.LinuxKernelCryptoAPI) ... ok

x86-64 Ubuntu 15.10 Skylake CPU 3.x is still blocking. It looks like I have to add another workaround for a Ubuntu quirk.

New changeset 55d77f5a7cb3 by Christian Heimes in branch 'default':
Issue 27744: test_aes_cbc is blocking x86-64 Ubuntu 15.10 Skylake CPU 3.x for a while. Require Kernel 4.3+ for now
https://hg.python.org/cpython/rev/55d77f5a7cb3

New changeset a951f8f30922 by Victor Stinner in branch 'default':
Cleanup socketmodule.c
https://hg.python.org/cpython/rev/a951f8f30922

New changeset 3a6917c73857 by Victor Stinner in branch 'default':
socket: Fix memory leak in sendmsg() and sendmsg_afalg()
https://hg.python.org/cpython/rev/3a6917c73857

What is the status of this issue? test_aead_aes_gcm() fails on my Fedora 25 (Python: default branch).

haypo@selma$ cat /etc/fedora-release 
Fedora release 25 (Twenty Five)
haypo@selma$ uname -r
4.9.5-200.fc25.x86_64

test test_socket failed -- Traceback (most recent call last):
  File "/home/haypo/prog/python/default/Lib/test/support/__init__.py", line 556, in wrapper
    return func(*args, **kw)
  File "/home/haypo/prog/python/default/Lib/test/test_socket.py", line 5515, in test_aead_aes_gcm
    res = op.recv(assoclen + len(plain) + taglen)
OSError: [Errno 22] Invalid argument

I'll look into the matter and push a fix after the migration to github today.

By the way problem with AES-GCM is tracked in https://bugs.python.org/issue29324 . It was caused in a Kernel API change. Jan has provided a fix. I need to find some spare time to dig into Kernel sources and verify the patch.