Richard Levitte [Thu, 28 Apr 2016 16:18:04 +0000 (18:18 +0200)]
VMS: only explicitely translate names in library C files.
When compiling all other C files, rely on the compiler to
automatically pick up the name translation information from the header
files __DECC_INCLUDE_{PRO,EPI}LOGUE.H.
Richard Levitte [Mon, 11 Apr 2016 16:42:52 +0000 (18:42 +0200)]
VMS: It seems DEC C doesn't handle certain header files quite right
With DEC C on VMS, you can use __DECC_INCLUDE_PROLOGUE.H and
__DECC_INCLUDE_EPILOGUE.H to include some DEC C specific features or
pragmas without having to touch the other header files.
It seems, however, that the current version of the compiler requires
the file names to be upcased, or it doesn't handle them quite right.
Add aliases for des-ede-ecb and des-ede3-ecb ciphers.
Currently we can get all block ciphers with
EVP_get_cipherbyname("<alg_name>-<block-mode-name>")
for example, by names "aes-128-ecb" or "des-ede-cbc".
I found a problem with des-ede-ecb and des-ede3-ecb ciphers as
they can be accessed only with names:
EVP_get_cipherbyname("des-ede")
EVP_get_cipherbyname("des-ede3")
It breaks the general concept.
In this patch I add aliases which allow to use names:
EVP_get_cipherbyname("des-ede-ecb")
EVP_get_cipherbyname("des-ede3-ecb")
in addition to the currently used names.
Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Rich Salz <rsalz@openssl.org>
Matt Caswell [Thu, 28 Apr 2016 14:12:37 +0000 (15:12 +0100)]
Client side CKE processing can double free on error
The tls_client_key_exchange_post_work() frees the pms on error. It also
calls ssl_generate_master_secret() which also free the pms. If an error
occurs after ssl_generate_master_secret() has been called then a double
free can occur.
BIO_free should call method->destroy before free'ing member fields
Reviewed-by: Tim Hudson <tjh@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1007)
Christian Heimes [Tue, 19 Apr 2016 19:11:30 +0000 (21:11 +0200)]
Add getters for X509_STORE and X509_OBJECT members
OpenSSL 1.1.0-pre5 has made some additional structs opaque. Python's ssl
module requires access to some of the struct members. Three new getters
are added:
int X509_OBJECT_get_type(X509_OBJECT *a);
STACK_OF(X509_OBJECT) *X509_STORE_get0_objects(X509_STORE *v);
X509_VERIFY_PARAM *X509_STORE_get0_param(X509_STORE *ctx);
Signed-off-by: Christian Heimes <cheimes@redhat.com> Reviewed-by: Rich Salz <rsalz@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org>
Matt Caswell [Wed, 27 Apr 2016 12:18:38 +0000 (13:18 +0100)]
Don't leak memory on error in cms_RecipientInfo_pwri_crypt
The cms_RecipientInfo_pwri_crypt() allocated an EVP_CIPHER_CTX but then
failed to free it in some error paths. By allocating it a bit later that
can be avoided.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Matt Caswell [Wed, 27 Apr 2016 11:59:19 +0000 (12:59 +0100)]
Don't leak memory on error in BN_generate_prime_ex
In BN_generate_prime_ex() we do some sanity checks first and return
with an error if they fail. We should do that *before* allocating any
resources to avoid a memory leak.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Richard Levitte [Thu, 28 Apr 2016 11:19:38 +0000 (13:19 +0200)]
Make BIO_sock_error return a proper error code when getsockopt fails
BIO_sock_error() returned 1 when getsockopt() fails when it should
return the error code for that failure.
Additionally, the optlen parameter to getsockopt() has to point at
the size of the area that the optval parameter points at rather than
zero. Some systems may forgive it being zero, but others don't.
The traditional private key encryption algorithm doesn't function
properly if the IV length of the cipher is zero. These ciphers
(e.g. ECB mode) are not suitable for private key encryption
anyway.
Viktor Dukhovni [Tue, 26 Apr 2016 18:17:57 +0000 (14:17 -0400)]
Future proof build_chain() in x509_vfy.c
Coverity reports a potential NULL deref when "2 0 0" DANE trust-anchors
from DNS are configured via SSL_dane_tlsa_add() and X509_STORE_CTX_init()
is called with a NULL stack of untrusted certificates.
Since ssl_verify_cert_chain() always provideds a non-NULL stack of
untrusted certs, and no other code path enables DANE, the problem
can only happen in applications that use SSL_CTX_set_cert_verify_callback()
to implement their own wrappers around X509_verify_cert() passing
only the leaf certificate to the latter.
Regardless of the "improbability" of the problem, we do need to
ensure that build_chain() handles this case correctly.
Richard Levitte [Mon, 25 Apr 2016 18:28:54 +0000 (20:28 +0200)]
RSA, DSA, DH: Allow some given input to be NULL on already initialised keys
The diverse {RSA,DSA,DH}_set0_* functions are made to allow some
parameters to be NULL IF the corresponding numbers in the given key
structure have already been previously initialised. Specifically,
this allows the addition of private components to be added to a key
that already has the public half, approximately like this:
RSA_get0_key(rsa, NULL, &e, NULL);
RSA_get0_factors(rsa, &p, &q);
/* calculate new d */
RSA_set0_key(rsa, NULL, NULL, d);
Benjamin Kaduk [Mon, 18 Apr 2016 17:50:34 +0000 (12:50 -0500)]
Remove dead-code infinite loop
Commit d32f5d8733df9938727710d4194e92813c421ef1 added a 'goto end;' statement
at the end of the code block for the 'end' label. Fortunately, it was after a
return statement, so no infinite loop occurred, but it is still dead code.
Remove the extra goto statement as cleanup.
Reviewed-by: Matt Caswell <matt@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org>
Andy Polyakov [Fri, 15 Apr 2016 14:30:29 +0000 (16:30 +0200)]
crypto/poly1305/asm: chase overflow bit on x86 and ARM platforms.
Even though no test could be found to trigger this, paper-n-pencil
estimate suggests that x86 and ARM inner loop lazy reductions can
loose a bit in H4>>*5+H0 step.
Richard Levitte [Thu, 21 Apr 2016 12:30:08 +0000 (14:30 +0200)]
Build system: add include directories and dependencies for generators
In the case of generating a file like this:
GENERATE[foo.S]=mkfoo.pl arg1 arg2
the 'mkfoo.pl' generator itself might need to include other files,
such as perl modules within our source tree. We can reuse already
existing syntax for it, like this:
INCLUDE[mkfoo.pl]=module/path
or:
DEPEND[mkfoo.pl]=modules/mymodule.pm
This change implements the support for such constructs, and for the
DEPEND statement, for any value that indicates a perl module (.pm
file), it will automatically infer an INCLUDE statement for its
directory, just like it does for C header files, so you won't have do
write this:
IBM argues that in certain scenarios capability query is really
expensive. At the same time it's asserted that query results can
be safely cached, because disabling CPACF is incompatible with
reboot-free operation.
Fix EC_KEY_set_private_key() to call key->group->meth->set_private()
Fix a bug introduced by 6903e2e7e9a4 (Extended EC_METHOD customisation
support., 2016-02-01). key->meth->set_private() is wrongly called where
it should call key->group->meth->set_private().
PR#4517
Reviewed-by: Tim Hudson <tjh@openssl.org> Reviewed-by: Stephen Henson <steve@openssl.org>
Harden ASN.1 BIO handling of large amounts of data.
If the ASN.1 BIO is presented with a large length field read it in
chunks of increasing size checking for EOF on each read. This prevents
small files allocating excessive amounts of data.
CVE-2016-2109
Thanks to Brian Carpenter for reporting this issue.
Richard Levitte [Thu, 21 Apr 2016 16:03:16 +0000 (18:03 +0200)]
Warn when doing an out-of-source build and finding in-source build artifacts
The reason to warn is that configuration *may* pick up on
configuration header files that are in the source tree, that might be
for a wildly different configuration than what is expected in the
current out-of-source configuration.
Viktor Dukhovni [Fri, 22 Apr 2016 00:00:58 +0000 (20:00 -0400)]
Enabled DANE only when at least one TLSA RR was added
It is up to the caller of SSL_dane_tlsa_add() to take appropriate
action when no records are added successfully or adding some records
triggers an internal error (negative return value).
With this change the caller can continue with PKIX if desired when
none of the TLSA records are usable, or take some appropriate action
if DANE is required.
Also fixed the internal ssl_dane_dup() function to properly initialize
the TLSA RR stack in the target SSL handle. Errors in ssl_dane_dup()
are no longer ignored.
Matt Caswell [Tue, 19 Apr 2016 22:33:35 +0000 (23:33 +0100)]
Don't set peer_tmp until we have finished constructing it
If we fail halfway through constructing the peer_tmp EVP_PKEY but we have
already stored it in s->s3->peer_tmp then if anything tries to use it then
it will likely fail. This was causing s_client to core dump in the
sslskewith0p test. s_client was trying to print out the connection
parameters that it had negotiated so far. Arguably s_client should not do
that if the connection has failed...but given it is existing functionality
it's easier to fix libssl.
Matt Caswell [Fri, 22 Apr 2016 11:10:19 +0000 (12:10 +0100)]
Fix no-dsa on Windows/VMS
The no-dsa option was failing on Windows because some symbols were not
correctly flagged in libcrypto.num. Problem found due to the new symbol
consistency test.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Richard Levitte [Fri, 22 Apr 2016 08:20:59 +0000 (10:20 +0200)]
Add a best effort test to check shared library consistency
Our main development platforms are of the Unix family, which doesn't
have the same strictness regarding a shared library being consistent
with the contents of the ld script (.map file, on Linux and Solaris)
as Windows is with the contents of the .def file or VMS is with the
linker symb_vector option.
To eliminate surprises, we therefore need to make sure to check that
the contents of the .map file is matched with the shared library, at
least to check that the shared library isn't missing any symbols that
should be present.
This test isn't absolutely perfect, as it will only check the symbols
that would be present on Linux / Solaris and will therefore miss those
that would only appear on Windows or VMS. On the other hand, those
platform specific are few and far apart in time, so it's not likely
they will pose a problem.
If allocation in CRYPTO_clear_realloc() fails don't free up the original
buffer: this is consistent with the behaviour of realloc(3) and is expected
in other places in OpenSSL.
Reviewed-by: Rich Salz <rsalz@openssl.org> Reviewed-by: Richard Levitte <levitte@openssl.org> Reviewed-by: Viktor Dukhovni <viktor@openssl.org>