BN_pseudo_rand() and BN_pseudo_rand_range() are deprecated in
OpenSSL 3.0. Since they are identical to their non-'pseudo' version
anyway, let's make them alias.
2d34e85ddf
SSLSocket#connect eventually calls `GetOpenFile` in order to get the
underlying file descriptor for the IO object passed in on
initialization. `GetOpenFile` assumes that the Ruby object passed in is
a T_FILE object and just casts it to a T_FILE without any checks. If
you pass an object that *isn't* a T_FILE to that function, the program
will segv.
Since we assume the IO object is a file in the `connect` method, this
commit adds a `CheckType` in the initialize method to ensure that the IO
object is actually a T_FILE. If the object *isn't* a T_FILE, this class
will segv on `connect`, so I think this is a backwards compatible
change.
919fa44ec2
OpenSSL 3.0 refuses to generate DSA parameters shorter than 2048 bits,
but generating 2048 bits parameters takes very long time. Let's use EC
in these test cases instead.
c732387ee5
A certificate can only have one SubjectAltName extension. OpenSSL 3.0
performs a stricter validation and certificates containing multiple SANs
will be rejected.
558cfbe5f5
OpenSSL's PKCS12_create() by default uses pbewithSHAAnd40BitRC2-CBC for
encryption of the certificates. However, in OpenSSL 3.0, the algorithm
is part of the legacy provider and is not enabled by default.
Specify another algorithm that is still in the default provider for
these test cases.
998406d18f
OpenSSL 3.0 slightly changed the error message for a certificate
verification failure when an untrusted self-signed certificate is found
in the chain.
b5a0a19850
Remove availability test for MD4 and RIPEMD160 as they are considered
legacy and may be missing depending on the compile-time options of
OpenSSL. OpenSSL 3.0 by default disables them.
a3e59f4c2e
Disabling ECC support of OpenSSL is impractical nowadays.
We still try to have the C extension compile on no-ec builds (as well
as no-dh or no-engine, etc.) as long as we can, but keeping test cases
for such an extreme scenario is not worth the effort.
2cd01d4676
It uses deprecated PKey::{RSA,DSA,DH}#set_* methods, which will not
work with OpenSSL 3.0. The same can easily be achieved using
PKey#public_to_der regardless of the key kind.
7b66eaa2db
Drop support for Ruby 2.3, 2.4, and 2.5.
As of 2021-10, Ruby 2.6 is the oldest version that still receives
security fixes from the Ruby core team, so it doesn't make much sense
to keep code for those ancient versions.
3436bd040d
[ This is a backport to the 2.1 branch. ]
Fixed misspellings reported at [Bug #16437], for default gems.
(cherry picked from commit ruby/ruby@e68999c82c)
0f43789503
Provide a wrapper of SSL_set0_tmp_dh_pkey()/SSL_CTX_set_tmp_dh(), which
sets the DH parameters used for ephemeral DH key exchange.
SSLContext#tmp_dh_callback= already exists for this purpose, as a
wrapper around SSL_CTX_set_tmp_dh_callback(), but it is considered
obsolete and the OpenSSL API is deprecated for future removal. There is
no practical use case where an application needs to use different DH
parameters nowadays. This was originally introduced to support export
grade ciphers.
RDoc for #tmp_dh_callback= is updated to recommend the new #tmp_dh=.
Note that current versions of OpenSSL support automatic ECDHE curve
selection which is enabled by default. SSLContext#tmp_dh= should only be
necessary if you must allow ancient clients which don't support ECDHE.
aa43da4f04
Also, OpenSSL::BN::CONSTTIME is added.
OpenSSL itself had a feature that was vulnerable against a side-channel
attack. The OpenSSL authors determined that it was not a security issue,
and they have already fixed the issue by using BN_set_flags.
https://github.com/openssl/openssl/pull/13888
If a Ruby OpenSSL user was faced with a similar issue, they couldn't
prevent the issue because Ruby OpenSSL lacks a wrapper to BN_set_flags.
For the case, this change introduces the wrapper.
1e565eba89
Add a variant of PKey#sign and #verify that do not hash the data
automatically.
Sometimes the caller has the hashed data only, but not the plaintext
to be signed. In that case, users would have to use the low-level API
such as RSA#private_encrypt or #public_decrypt directly.
OpenSSL 1.0.0 and later supports EVP_PKEY_sign() and EVP_PKEY_verify()
which provide the same functionality as part of the EVP API. This patch
adds wrappers for them.
16cca4e0c4
Methods that take both PEM-encoding and DER-encoding have not been
consistent in the order in which encoding to attempt to parse.
A DER-encoding may contain a valid PEM block ("\n-----BEGIN ..-----" to
"-----END ...-----") embedded within it. Also, the PEM-encoding parser
allows arbitrary data around the PEM block and silently skips it. As a
result, attempting to parse data in DER-encoding as PEM-encoding first
can incorrectly finds the embedded PEM block instead.
This commit ensures that DER encoding will always be attempted before
PEM encoding. OpenSSL::X509::Certificate is one of the updated classes.
With this, the following will always be true:
# obj is an OpenSSL::X509::Certificate
obj == OpenSSL::X509::Certificate.new(obj.to_der)
obj == OpenSSL::X509::Certificate.new(obj.to_pem)
b280eb1fd0
Normal sockets respond to `getbyte`, so we should make SSLSocket respond
to `getbyte` as well. This way we can substitute SSLSockets for regular
sockets.
ac1490b7c9
Use EVP_PKEY_param_check() instead of DH_check() if available. Also,
use EVP_PKEY_public_check() instead of EC_KEY_check_key().
EVP_PKEY_*check() is part of the EVP API and is meant to replace those
low-level functions. They were added by OpenSSL 1.1.1. It is currently
not provided by LibreSSL.
797e9f8e08
The low-level API that is used to implement #public_key is deprecated
in OpenSSL 3.0. It is actually very simple to implement in another way,
using existing methods only, in much shorter code. Let's do it.
While we are at it, the documentation is updated to recommend against
using #public_key. Now that OpenSSL::PKey::PKey implements public_to_der
method, there is no real use case for #public_key in newly written Ruby
programs.
48a6c391ef
Use EVP_PKEY_print_private() instead of the low-level API *_print()
functions, such as RSA_print().
EVP_PKEY_print_*() family was added in OpenSSL 1.0.0.
Note that it falls back to EVP_PKEY_print_public() and
EVP_PKEY_print_params() as necessary. This is required for EVP_PKEY_DH
type for which _private() fails if the private component is not set in
the pkey object.
Since the new API works in the same way for all key types, we now
implement #to_text in the base class OpenSSL::PKey::PKey rather than in
each subclass.
e0b4c56956
Implement PKey::DSA.new(size) and PKey::DSA.generate using
OpenSSL::PKey.generate_parameters and .generate_key instead of the low
level DSA functions.
1800a8d5eb
Implement PKey::DH.new(size, gen), PKey::DH.generate(size, gen), and
PKey::DH#generate_key! using PKey.generate_parameters and .generate_key
instead of the low level DH functions.
Note that the EVP interface can enforce additional restrictions - for
example, DH key shorter than 2048 bits is no longer accepted by default
in OpenSSL 3.0. The test code is updated accordingly.
c2e9b16f0b
Similarly to OpenSSL::PKey.generate_key and .generate_parameters, let
OpenSSL::PKey::PKey#sign and #verify take an optional parameter for
specifying control strings for EVP_PKEY_CTX_ctrl_str().
faf85d7c1d
Clean up old version guards in preparation for the upcoming OpenSSL 3.0
support.
OpenSSL 1.0.1 reached its EOL on 2016-12-31. At that time, we decided
to keep 1.0.1 support because many major Linux distributions were still
shipped with 1.0.1. Now, nearly 4 years later, most Linux distributions
are reaching their EOL and it should be safe to assume nobody uses them
anymore. Major ones that were using 1.0.1:
- Ubuntu 14.04 is EOL since 2019-04-30
- RHEL 6 will reach EOL on 2020-11-30
LibreSSL 3.0 and older versions are no longer supported by the LibreSSL
team as of October 2020.
Note that OpenSSL 1.0.2 also reached EOL on 2019-12-31 and 1.1.0 also
did on 2018-08-31.
c055938f4b
ASN1_dup() will not copy the 'pkey' field of a PKCS7_SIGNER_INFO object
by design; it is a temporary field kept until the PKCS7 structure is
finalized. Let's bump reference counter of the pkey in the original
object and use it in the new object, too.
This commit also removes PKCS7#add_signer's routine to add the
content-type attribute as a signed attribute automatically. This
behavior was not documented or tested. This change should not break any
working user code since the method was completely useless without the
change above.
20ca7a27a8
The .include directive was initially added by OpenSSL 1.1.1, but the
syntax was later modified in 1.1.1b to improve compatibility with the
parser in <= 1.1.0. The test case expects 1.1.1b's parser.
95f59d398c
The test case is failing on Ubuntu 18.04 because it still uses the
initial 1.1.1 release:
20210316T120003Z.fail.html.gz
AES CCM mode in OpenSSL <= 1.1.1b was overly strict in the parameters
assignment order. This has been relaxed by OpenSSL 1.1.1c.
b48e3be947
The test case is failing on Ubuntu 18.04 because it still uses the
initial 1.1.1 release and has the issue:
20210316T120003Z.fail.html.gz
LibreSSL 3.2.4 made the certificate verification logic back closer to
pre-3.2.2 one, which is more compatible with OpenSSL.
Part of the fixes added by commit a0e98d48c91f ("Enhance TLS 1.3 support
on LibreSSL 3.2/3.3", 2020-12-03) is required for 3.2.2 and 3.2.3 only
(and ~3.3.1, however 3.3 does not have a stable release yet). Since both
releases are security fix, it should be safe to remove those special
treatment from our test suite.
While we are at it, TestSSL#test_ecdh_curves is split into TLS 1.2 and
TLS 1.3 variants for clarity.
a9954bac22
This defines TLS1_3_VERSION when using LibreSSL 3.2+. LibreSSL 3.2/3.3
doesn't advertise this by default, even though it will use TLS 1.3
in both client and server modes.
Changes between LibreSSL 3.1 and 3.2/3.3 broke a few tests, Defining
TLS1_3_VERSION by itself fixes 1 test failure. A few tests now
fail on LibreSSL 3.2/3.3 unless TLS 1.2 is set as the maximum version,
and this adjusts those tests. The client CA test doesn't work in
LibreSSL 3.2+, so I've marked that as pending.
For the hostname verification, LibreSSL 3.2.2+ has a new stricter
hostname verifier that doesn't like subjectAltName such as
c*.example.com and d.*.example.com, so adjust the related tests.
With these changes, the tests pass on LibreSSL 3.2/3.3.
a0e98d48c9
The underlying API SSL_CTX_set_tmp_ecdh_callback() was removed by
LibreSSL >= 2.6.1 and OpenSSL >= 1.1.0, in other words, it is not
supported by any non-EOL versions of OpenSSL.
The wrapper was initially implemented in Ruby 2.3 and has been
deprecated since Ruby/OpenSSL 2.0 (bundled with Ruby 2.4) with explicit
warning with rb_warn().
ee037e1460
Undo special treatment of nil and simply pass the value to
StringValueCStr().
nil was never a valid argument for the method; OpenSSL::X509::StoreError
with an unhelpful error message "system lib" was raised in that case.
fb2fcbb137
SSLContext's verify_mode expects an SSL_VERIFY_* constant (an integer)
and verify_hostname expects either true or false. However, they are set
to nil after calling OpenSSL::SSL::SSLContext.new, which is surprising.
Set a proper value to them by default: verify_mode is set to
OpenSSL::SSL::VERIFY_NONE and verify_hostname is set to false by
default.
Note that this does not change the default behavior. The certificate
verification was never performed unless verify_mode is set to
OpenSSL::SSL::VERIFY_PEER by a user. The same applies to
verify_hostname.
87d869352c
Add explicit test cases for the behaviors with different verify_mode.
If we made a bug in verify_mode, we would notice it by failures of other
test cases, but there were no dedicated test cases for verify_mode.
1ccdc05662
The current test_client_auth_public_key test case checks that supplying
a PKey containing only public components through client_cert_cb will
cause handshake to fail. While this is a correct behavior as a whole,
the assertions are misleading in the sense that giving a public key is
causing the failure. Actually, the handshake fails because a client
certificate is not supplied at all, as a result of ArgumentError that is
silently ignored.
Rename the test case to test_client_cert_cb_ignore_error and simplify it
to clarify what it is testing.
785b5569fc
Explicitly check for type given some conflicting statements within openssl's
documentation around EVP_PKEY_cmp and EVP_PKEY_ASN1_METHOD(3).
Add documentation with an example for compare?
0bf51da6e2
