Matt Caswell [Thu, 25 Jun 2015 08:47:15 +0000 (09:47 +0100)]
Reject calls to X509_verify_cert that have not been reinitialised
The function X509_verify_cert checks the value of |ctx->chain| at the
beginning, and if it is NULL then it initialises it, along with the value
of ctx->untrusted. The normal way to use X509_verify_cert() is to first
call X509_STORE_CTX_init(); then set up various parameters etc; then call
X509_verify_cert(); then check the results; and finally call
X509_STORE_CTX_cleanup(). The initial call to X509_STORE_CTX_init() sets
|ctx->chain| to NULL. The only place in the OpenSSL codebase where
|ctx->chain| is set to anything other than a non NULL value is in
X509_verify_cert itself. Therefore the only ways that |ctx->chain| could be
non NULL on entry to X509_verify_cert is if one of the following occurs:
1) An application calls X509_verify_cert() twice without re-initialising
in between.
2) An application reaches inside the X509_STORE_CTX structure and changes
the value of |ctx->chain| directly.
With regards to the second of these, we should discount this - it should
not be supported to allow this.
With regards to the first of these, the documentation is not exactly
crystal clear, but the implication is that you must call
X509_STORE_CTX_init() before each call to X509_verify_cert(). If you fail
to do this then, at best, the results would be undefined.
Calling X509_verify_cert() with |ctx->chain| set to a non NULL value is
likely to have unexpected results, and could be dangerous. This commit
changes the behaviour of X509_verify_cert() so that it causes an error if
|ctx->chain| is anything other than NULL (because this indicates that we
have not been initialised properly). It also clarifies the associated
documentation. This is a follow up commit to CVE-2015-1793.
Matt Caswell [Wed, 24 Jun 2015 14:55:36 +0000 (15:55 +0100)]
Fix alternate chains certificate forgery issue
During certificate verfification, OpenSSL will attempt to find an
alternative certificate chain if the first attempt to build such a chain
fails. An error in the implementation of this logic can mean that an
attacker could cause certain checks on untrusted certificates to be
bypassed, such as the CA flag, enabling them to use a valid leaf
certificate to act as a CA and "issue" an invalid certificate.
This occurs where at least one cert is added to the first chain from the
trust store, but that chain still ends up being untrusted. In that case
ctx->last_untrusted is decremented in error.
The PSK identity hint should be stored in the SSL_SESSION structure
and not in the parent context (which will overwrite values used
by other SSL structures with the same SSL_CTX).
Rewrite ssl3_digest_cached_records handling. Only digest cached records
if digest array is NULL: this means it is safe to call
ssl3_digest_cached_records multiple times (subsequent calls are no op).
Remove flag TLS1_FLAGS_KEEP_HANDSHAKE instead only update handshake buffer
if digest array is NULL.
Add additional "keep" parameter to ssl3_digest_cached_records to indicate
if the handshake buffer should be retained after digesting cached records
(needed for TLS 1.2 client authentication).
Add secure heap for storage of private keys (when possible).
Add BIO_s_secmem(), CBIGNUM, etc.
Add BIO_CTX_secure_new so all BIGNUM's in the context are secure.
Contributed by Akamai Technologies under the Corporate CLA.
Reviewed-by: Richard Levitte <levitte@openssl.org>
As numerous comments indicate the certificate and key array is not an
appopriate structure to store the peers certificate: so remove it and
just the s->session->peer instead.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Richard Levitte [Sun, 21 Jun 2015 17:11:43 +0000 (19:11 +0200)]
Cleanup mttest.c : modernise output
Construct bio_err and bio_stdout from file handles instead of FILE
pointers, since the latter might not be implemented (when OPENSSL_NO_STDIO
is defined).
Convert all output to use BIO_printf.
Change lh_foo to lh_SSL_SESSION_foo.
Rich Salz [Sat, 2 May 2015 14:01:33 +0000 (10:01 -0400)]
RT2547: Tighten perms on generated privkey files
When generating a private key, try to make the output file be readable
only by the owner. Put it in CHANGES file since it might be noticeable.
Add "int private" flag to apps that write private keys, and check that it's
set whenever we do write a private key. Checked via assert so that this
bug (security-related) gets fixed. Thanks to Viktor for help in tracing
the code-paths where private keys are written.
Rich Salz [Sat, 13 Jun 2015 14:50:00 +0000 (10:50 -0400)]
Refactor into clear_ciphers; RT3588
While closing RT3588 (Remove obsolete comment) Kurt and I saw that a
few lines to completely clear the SSL cipher state could be moved into
a common function.
Emilia Kasper [Tue, 9 Jun 2015 12:17:50 +0000 (14:17 +0200)]
Remove SSL_OP_TLS_BLOCK_PADDING_BUG
This is a workaround so old that nobody remembers what buggy clients
it was for. It's also been broken in stable branches for two years and
nobody noticed (see
https://boringssl-review.googlesource.com/#/c/1694/).
It should not be possible for DTLS message fragments to span multiple
packets. However previously if the message header fitted exactly into one
packet, and the fragment body was in the next packet then this would work.
Obviously this would fail if packets get re-ordered mid-flight.
Matt Caswell [Thu, 4 Jun 2015 13:22:00 +0000 (14:22 +0100)]
EC_POINT_is_on_curve does not return a boolean
The function EC_POINT_is_on_curve does not return a boolean value.
It returns 1 if the point is on the curve, 0 if it is not, and -1
on error. Many usages within OpenSSL were incorrectly using this
function and therefore not correctly handling error conditions.
With thanks to the Open Crypto Audit Project for reporting this issue.
Matt Caswell [Thu, 4 Jun 2015 09:35:08 +0000 (10:35 +0100)]
Change BIO_number_read and BIO_number_written() to be 64 bit
The return type of BIO_number_read() and BIO_number_written() as well as
the corresponding num_read and num_write members in the BIO structure has
been changed from unsigned long to uint64_t. On platforms where an unsigned
long is 32 bits (e.g. Windows) these counters could overflow if >4Gb is
transferred.
With thanks to the Open Crypto Audit Project for reporting this issue.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Matt Caswell [Thu, 30 Apr 2015 14:20:25 +0000 (15:20 +0100)]
Tighten extension handling
This adds additional checks to the processing of extensions in a ClientHello
to ensure that either no extensions are present, or if they are then they
take up the exact amount of space expected.
With thanks to the Open Crypto Audit Project for reporting this issue.
Matt Caswell [Thu, 30 Apr 2015 13:51:10 +0000 (14:51 +0100)]
Fix memory leaks in BIO_dup_chain()
This fixes a memory leak that can occur whilst duplicating a BIO chain if
the call to CRYPTO_dup_ex_data() fails. It also fixes a second memory leak
where if a failure occurs after successfully creating the first BIO in the
chain, then the beginning of the new chain was not freed.
With thanks to the Open Crypto Audit Project for reporting this issue.
Matt Caswell [Thu, 30 Apr 2015 13:04:30 +0000 (14:04 +0100)]
Replace memset with OPENSSL_clear_free()
BUF_MEM_free() attempts to cleanse memory using memset immediately prior
to a free. This is at risk of being optimised away by the compiler, so
replace with a call to OPENSSL_clear_free() instead.
With thanks to the Open Crypto Audit Project for reporting this issue.
Richard Levitte [Tue, 9 Jun 2015 21:06:23 +0000 (23:06 +0200)]
When making libcrypto from apps or test, make sure to include engines
For librypto to be complete, the stuff in both crypto/ and engines/
have to be built. Doing 'make test' or 'make apps' from a clean
source tree failed to do so.
Corrected by using the new 'build_libcrypto' in the top Makefile.
Richard Levitte [Tue, 9 Jun 2015 23:34:26 +0000 (01:34 +0200)]
Add and rearrange building of libraries
There's a need for a target that will build all of libcrypto, so let's
add 'build_libcrypto' that does this. For ortogonality, let's also
add 'build_libssl'. Have both also depend on 'libcrypto.pc' and
'libssl.pc' so those get built together with the libraries.
This makes 'all' depend on fewer things directly.
Check return value when calling ASN1_INTEGER_get to retrieve a certificate
serial number. If an error occurs (which will be caused by the value being
out of range) revert to hex dump of serial number.