OSSL_STORE's loading function could prompt repeatedly for the same
passphrase. It turns out that OSSL_STORE_load() wasn't caching the
passphrase properly. Fixed in this change.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15064)
In the provider file: scheme loader implementation, the OSSL_DECODER_CTX
was set up with all sorts of implementations, even if the caller has
declared a limited expectation on what should be loaded, which means
that even though a certificate is expected, all the diverse decoders
to produce an EVP_PKEY are added to the decoding change.
This optimization looks more closely at the expected type, and only
adds the EVP_PKEY related decoder implementations to the chain if
there is no expectation, or if the expectation is one of
OSSL_STORE_INFO_PARAMS, OSSL_STORE_INFO_PUBKEY, OSSL_STORE_INFO_PKEY.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15066)
A pair of the disabled string checks were incorrect.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/15082)
A newly created BIO object within this function calls
OSSL_STORE_attach() which increases the ref count to 2.
OSSL_STORE_close() then decrements the ref count by 1, so the BIO still
remains.
The following new test was picking up this leak using..
> valgrind openssl crl -hash -noout < test/testcrl.pem
Not quite sure why the existing tests were not picking this up
since they appear to run through a similiar path.. such as
> valgrind openssl pkey < test-runs/test_rsa/rsa-pkcs8-ff.dd
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15058)
Fixes#15031
The maybe_stdin needed to be passed to load_key_certs_crls().
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15058)
The EVP_PKEY_asn1_set_public and EVP_PKEY_meth_set_copy have
some API breaking constification changes in 3.0.
Fixes#9296
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15068)
Just something I noticed while reading this code.
This was probably committed a very long time ago.
Fixed typo anther -> another.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15078)
SHA1 is now in security level 0. SHA1 is required for DTLSv1.1. Therefore
ssl_test_new needed some adjustments in the event that DTLSv1.2 is disabled.
There are also adjustments required if using the FIPS module and DTLSv1.2
is disabled. The only DTLS version supported by the FIPS module is
DTLSv1.2.
Fixes #14956
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15047)
SHA1 is now in security level 0. SHA1 is required for DTLSv1.1. Therefore
sslapitest needed some adjustments in the event that DTLSv1.2 is disabled.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15047)
SHA1 is now in security level 0. SHA1 is required for DTLSv1.1. Therefore
dtlstest needed some adjustments in the event that DTLSv1.2 is disabled.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15047)
ASN1_put_object() was preceded by the nonsensical comment:
/*
* class 0 is constructed constructed == 2 for indefinite length constructed
*/
This is the result of concatenating two sentences in 0f113f3ee4 by
automated reformatting. The first sentence, "class 0 is constructed",
goes back to d02b48c63a, the import of SSLeay 0.8.1b. Even in that
context, it made little sense; class 0 means "universal", not
constructed, and there is no special significance to class 0 in this
function in any case.
Therefore I have simply removed that first sentence.
CLA: trivial
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/14998)
Note that configuration and installation procedure has changed:
- The FIPS provider is now disabled by default and needs to
be enabled by configuring with `enable-fips`.
- If the FIPS provider is enabled, it gets installed automatically.
There is no extra installation step required anymore.
This is more natural and coincides with the expectation of the
user, namely "what's configured, gets installed".
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
Building the fips provider in addition to the default provider
effectively doubles the build time. Since many users will not
need fips support, it is now disabled by default.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
Pull request #14320 introduced the ability to use compiled programs
as generators in GENERATE rules of build.info files. Those generator
calls were wrapped by the Perl wrapper (wrap.pl) in the Unix makefile
template, but not on Windows.
This commit adds the missing wrapper for Windows, because for the
`fipsmodule.cnf` target it is essential that the `openssl fipsinstall`
command does not load any preinstalled openssl configuration file.
Fixes#13680
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
To follow the principle "what you configure is what you install",
the `make install` target now includes the installation of the
fips provider (`make install_fips`) if (and only if) OpenSSL was
configured with fips support (`enable-fips`).
The `make install_fips` target exists as well and can be used
to install just the fips provider. It requires `enable-fips`
and issues an error message if `no-fips` was configured.
The anologue holds for the 'uninstall_fips' target.
Fixes#13693
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
There is already a `providers/fipsmodule.cnf` target which is required by
the tests. Instead of creating another fipsmodule.cnf, the `install_fips`
target simply copies that configuration file to its final destination.
This commit also restores the minimal dependencies to build the `install_fips`
target immediately after configuring, which was broken after the removal
of the `install_sw` dependency.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
According to the OpenSSL 3.0 Wiki, the file should be located at
$(DESTDIR)$(OPENSSLDIR)/fipsmodule.cnf
next to the openssl.cnf file.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
The `openssl` app was previously called without a path, which
would generally invoke the system's copy of the openssl application.
Currently, that's most likely an openssl version 1.1.1 application,
which does not recognize the `fipsinstall` command and terminates
with an error message.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
The FIPS hmac key is provided as a hexadezimal string, which needs to
be be prefixed with `hexkey:`, not `key:`.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/13684)
Fixes#15056
The dependency for fipsinstall was being added to the makefile regardless of
it being used. This means that a subsequent `make test` would fail if the
command line application wasn't present. Rather than fix the instance in question,
it is better to leave out this part of the makefile if the tests cannot be
run.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15057)
It was discovered that MSVC has localized /showIncludes output.
Fortunately, it still seems to follow the same generic format, so we
can adapt the regular expression to make it language agnostic.
Fixes#14994
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/15006)
Previously, we had dependency making pretty much hard coded in the
build file templates, with a bit of an exception for Unix family
platforms, where we had different cases depending on what dependency
making program was found.
With the Embarcadero C++ builder, a separate scheme appeared, with a
different logic.
This change merges the two, and introduces two config target
attributes:
makedepcmd The program to use, where this is relevant.
This replaces the earlier configuration
attribute 'makedepprog'.
makedep_scheme This is a keyword that can be used by build
files templates to produce different sorts of
commands, but most importantly, to pass as
argument to util/add-depend.pl, which uses
this keyword as a "producer" for the
dependency lines.
If the config target doesn't define the 'makedep_scheme' attribute,
Configure tries to figure it out by looking for GCC compatible
compilers or for the 'makedepend' command.
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/15006)
The modified way to generate .d files had an unfortunate side effect,
that it outputs the whole preprocessed file and not just the dependency
lines, at least with MSVC's cl. That gave util/add-depends.pl a whole
lot more to read through, which impacts greatly on the performance of
dependency treatment.
We modify the process by adding a config target attribute 'make_depend',
which can be any suitable command for generating such lines. All it
needs is to also accept C flags and macro definitions.
Fixes#14994
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/15006)
We test that sending a Finished message instead of a ClientHello results
in an unexpected message error.
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14930)
In TLS we process received messages like this:
1) Read Message Header
2) Validate and transition state based on received message type
3) Read Message Body
4) Process Message
In DTLS we read messages like this:
1) Read Message Header and Body
2) Validate and transition state based on received message type
3) Process Message
The difference is because of the stream vs datagram semantics of the
underlying transport.
In both TLS and DTLS we were doing finished MAC processing as part of
reading the message body. This means that in DTLS this was occurring
*before* the state transition has been validated. A crash was occurring
in DTLS if a Finished message was sent in an invalid state due to
assumptions in the code that certain variables would have been setup by
the time a Finished message arrives.
To avoid this problem we shift the finished MAC processing to be after
the state transition in DTLS.
Thanks to github user @bathooman for reporting this issue.
Fixes#14906
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14930)
The provider config module was storing the list of activated providers
in a global variable. However, because different libctxs can each load
providers via config files we need to keep the list of activated providers
separate and in the libctx.
Partially fixes#15030
Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15033)
Check that we don't see any threading issues when loading/unloading a
provider from multiple threads.
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15010)
The X509 version APIs return the numerical values of the version
numbers, which are one off from the names. This is a bit confusing.
Where they don't get it wrong (accidentally making an "X509v4"
certificate), callers tend to try commenting every call site to explain
the mismatch, including in OpenSSL itself.
Define constants for these values, so code can be self-documenting and
callers are nudged towards the right values.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14549)
Disabling memleaktest under MSVC because leak detection is not a supported feature with MSVC's AddressSanitizer. Leaving ASan enabled in this case causes a test failure because the test suite is expecting the leak to be detected.
CLA: trivial
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14581)
It is possible for the stack of X509_OBJECTs held in an X509_STORE_CTX to
have a custom compare function associated with it. Normally (by default)
this uses X509_NAME_cmp(). The X509_STORE_CTX_get1_issuer() function
assumed that it would always be X509_NAME_cmp().
By implementing OPENSSL_sk_find_all() function we can avoid explicitly
using X509_NAME_cmp() in X509_STORE_CTX_get1_issuer().
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/14728)
