* support defining which windows libraries to link with an env var
CRYPTOGRAPHY_WINDOWS_LIBRARIES is your new friend
* add some docs
* change to CRYPTOGRAPHY_WINDOWS_LINK_OPENSSL110
* lib prefixing is not a thing msvc does, right
* enforce password must be bytes when loading PEM/DER asymmetric keys
Previously we were using an ffi.buffer on the Python string, which was
allowing text implicitly, but our documentation explicitly requires
bytes.
* add changelog entry
* DH keys support serialization
* Add DH serialization documentation
* Add tests for DH keys serialization in DER encoding
* update version to 1.8
* Allow only SubjectPublicKeyInfo serialization
* Remove support in TraditionalOpenSSL format
* Fix pep8
* Refactor dh serialization tests
* Backport DH_check from OpenSSL 1.1.0.
OpenSSL 1.0.2's DH_check considers the q parameter, allowing it
validate more generators and primes; however, OpenSSL 1.1.0's DH_check
includes code to handle errors in BN functions, so it's preferred.
* Wrap DH_Check when using OpenSSL 1.1.0 or higher.
* Adding DH_CHECK_* values missing from older OpenSSLs
* Defensively guard DH_CHECK_* definitions with ifndef.
This will prevent duplicate definitions when LibreSSL supports a
version of DH_check that can return these.
* Document the OpenSSL of origin for the DH_check code
* Use static callbacks with Python 3.x again
Static callbacks were disabled for Python 3.5+ to work around an issue
with subinterpreters, locking callbacks and osrandom engine. Locking
callback and osrandom engine were replaced with a C implementations in
version 1.6 and 1.7.
https://github.com/pyca/cryptography/issues/2970Closes: #3348
Signed-off-by: Christian Heimes <christian@python.org>
* remove unused import
* add DTLSv1_2 methods
* add binding to DTLSv1_get_timeout() and DTLSv1_handle_timeout()
* fix: PEP8 failed
fix the following error:
./src/_cffi_src/openssl/ssl.py:728:80: E501 line too long (80 > 79 characters)
see https://jenkins.cryptography.io/job/cryptography-pr-pep8/1954/
* Revert "add DTLSv1_2 methods"
This reverts commit e4a9150b12ddb4790159a5835f1d1136cb1b996e.
* replace 'long int' by 'long'
To be more consistent with the naming convention
cf 8dde92aad5 (r90153970)
* wrap with braces
cf 8dde92aad5 (r90154057)
* conditionally bind all DTLS
* rebase error
* rename wrapped function
* delete the 1.0.0 support
* drop the version check
* drop the AES-CTR stuff
* Update the example
* openssl truncates for us now
* delete unused test
* unused imports
* Remove a bunch of conditional bindings for NPN
* no more 1.0.0 builders
* libressl fix
* update the docs
* remove dead branches
* oops
* this is a word, damnit
* spelling
* try removing this
* this test is not needed
* unused import
* Scrypt docs code example contradict RFC 7914 (#3302)
* More secure example difficulty of parameter n in scrypt docs (#3302)
* Change link text to scrypt paper (#3302)
* Change link text to scrypt paper, part deux (#3302)
* Add "logins" to spelling wordlist
* New osrandom_engine in C
Inspired by Python/random.c and the old implementation.
Signed-off-by: Christian Heimes <christian@python.org>
* osrandom_engine
* Fix naming bug caused by search 'n replace mistake
* Make it easier to override osrandom auto-detection
* Add engine ctrl and backend API to get implementation from ENGINE
Signed-off-by: Christian Heimes <christian@python.org>
* Better test coverage, documentation, LICENSE
Signed-off-by: Christian Heimes <christian@python.org>
* Coverage is hard.
Signed-off-by: Christian Heimes <christian@python.org>
* * enable win32 check
* read() returns size_t
Signed-off-by: Christian Heimes <christian@python.org>
* Add macOS to spelling list. Remove dead code from header file.
Signed-off-by: Christian Heimes <christian@python.org>
* remove CCRandomGenerateBytes path and update getentropy to work on macOS
This change allows us to test all the engines in our CI:
* getentropy (tested by macOS sierra)
* getrandom (tested on several linux builders)
* /dev/urandom (tested on FreeBSD, OS X 10.11 and below, & older linux)
* CryptGenRandom (tested on windows builders)
I also fixed bugs preventing compilation in the getentropy code
* getentropy() returns int and is restricted to 256 bytes on macOS, too.
Signed-off-by: Christian Heimes <christian@python.org>
* add versionadded
* Re-add import of os module
* Fixes related to Alex's recent review.
Signed-off-by: Christian Heimes <christian@python.org>
* Add error reporting and fail for EAGAIN
Add error reporting strings for various error cases. This gives us much
nicer and understandable error messages.
SYS_getrandom() EAGAIN is now an error. Cryptography refuses to
initialize its osrandom engine when the Kernel's CPRNG hasn't been
seeded yet.
Signed-off-by: Christian Heimes <christian@python.org>
Cryptography uses new features from tox 2.4. Tox 2.3 happily ignores the
new config stanzes and doesn't install dependency. This can lead to strange
test failures.
With minversion=2.4, tox 2.3 fails to run properly:
$ tox
ERROR: tox version is 2.3.1, required is at least 2.4
Signed-off-by: Christian Heimes <christian@python.org>
* add cffi bindings to objects.py and evp.py (required for pypy's _hashlib implementation)
* ah, that comes from copying it from the man page
* dont use #define ..., delcare it as static cont long <name>
* Start of OpenSSL DH backend implementation
* Supporting DH in MultiBackend
* DHBackend has dh_parameters_supported method
* Removed DHParametersWithNumbers and DHPrivateKeyWithNumbers from documentation
* Removed ExchangeContext. exchange is a method of DHPrivateKeyWithSerialization
* PEP8 fixes
* Fixed TestDH.test_bad_tls_exchange
* Fixed generate_private_key reference in dh documentation
* test DH multibackend support
* testing DH coversion to serialized
* Validating that we receive serialized class in test_generate_dh
* Testing DH exchange symmetric key padding
* struct DH is now opaqued
* PEP8 fixes
* Testing load_dh_private_numbers throws ValueError when DH_check fails
* Using openssl_assert
* Passing keywords arguments in DH key exchange example
* test_dh::test_bad_tls_exchange now uses pre calculated parameters
* TestDH - Add test that the computed secret is equivalent to the definition by comparing with secret computed in pure python
* Add missing generator parameter to DHBackend interface docs.
* Include parameter type in DHBackend abc docs.
* Add docs for dh.generate_parameters function
* Remove the dh Numbers section, and move the DHNumbers class docs to where they are first used.
* Add note of big endian byte packing to DH exchange method.
* DH documentation updates.
Add single sentence overview with wikipedia link.
Add paragraph on assembling using Numbers objects.
Add link to backend interface docs.
First section was all indented, I think by mistake.
* Add exchange method to DHPrivateKey abstract base class.
* Small tweaks to DH documentation - remove Provider.
* Add endian to dictionary
* Use utils.int_from_bytes in test_tls_exchange_algorithm
* Removed duplicate line
* Change dh.rst exchange algorithm from doctest to code-block
The example in the Diffie-Hellman exhange algorithm is using
2048 bits key. Generating the parameters of 2048 takes long
time. This caused the automated tests to fail. In order to
pass the tests we change the example to code-block so it
will not run in the doc tests.
* Fix dh docs
* Document the generator in DHBackend relevant methods
* Fix dh tests
* use DHparams_dup
* Fix key type to unsigned char as expected by DH_compute_key
* Validate that DH generator is 2 or 5
* test dh exchange using botan vectors
* group all numbers classes
* Simplify _DHPrivateKey
* Rename test with serialized to numbers
* Move bad exchange params to external vector file
* update exchange versionadded to 1.7
* Make key_size bit accurate
* Change botan link
* Added CHANGELOG entry
