Move s->rstate to s->rlayer.rstate

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move s->packet and s->packet_length into s->rlayer

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove unneccessary use of accessor function now code is moved into record
layer

Reviewed-by: Richard Levitte <levitte@openssl.org>

Make rrec, wrec, rbuf and wbuf fully private to the record layer. Also, clean
up some access to them. Now that various functions have been moved into the
record layer they no longer need to use the accessor macros.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Removed dependency on rrec from heartbeat processing

Reviewed-by: Richard Levitte <levitte@openssl.org>

Introduce macro RECORD_LAYER_setup_comp_buffer

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix bug where rrec was being released...should have been removed by one of
the earlier record layer commits

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move ssl3_pending into the record layer

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove RECORD_LAYER_set_ssl and introduce RECORD_LAYER_init

Reviewed-by: Richard Levitte <levitte@openssl.org>

Provide RECORD_LAYER_set_data function

Reviewed-by: Richard Levitte <levitte@openssl.org>

Introduce the functions RECORD_LAYER_release, RECORD_LAYER_read_pending, and
RECORD_LAYER_write_pending.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Create RECORD_LAYER_clear function.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Tidy up rec_layer.h. Add some comments regarding which functions should be
being used for what purpose.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Moved s3_pkt.c, s23_pkt.c and d1_pkt.c into the record layer.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Split out non record layer functions out of s3_pkt.c and d1_pkt.c into
the new files s3_msg.c and s1_msg.c respectively.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move more SSL3_RECORD oriented functions into ssl3_record.c

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move SSL3_RECORD oriented functions into ssl3_record.c

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move SSL3_BUFFER set up and release code into ssl3_buffer.c

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move s->s3->wrec to s>rlayer>wrec

Reviewed-by: Richard Levitte <levitte@openssl.org>

Encapsulate s->s3->wrec

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move s->s3->rrec to s->rlayer->rrec

Reviewed-by: Richard Levitte <levitte@openssl.org>

Encapsulate s->s3->rrec

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move s->s3->wbuf to s->rlayer->wbuf

Reviewed-by: Richard Levitte <levitte@openssl.org>

Encapsulate access to s->s3->wbuf

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move s->s3->rrec into s->rlayer

Reviewed-by: Richard Levitte <levitte@openssl.org>

Encapsulate SSL3_BUFFER and all access to s->s3->rbuf.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Create a RECORD_LAYER structure and move read_ahead into it.

Reviewed-by: Richard Levitte <levitte@openssl.org>

update ordinals

Reviewed-by: Matt Caswell <matt@openssl.org>

Move more internal only functions to asn1_locl.h

Reviewed-by: Matt Caswell <matt@openssl.org>

free NULL cleanup.

This gets EC_GROUP_clear_free EC_GROUP_free, EC_KEY_free,
EC_POINT_clear_free, EC_POINT_free

Reviewed-by: Kurt Roeckx <kurt@openssl.org>

Resolve swallowed returns codes

The recent updates to libssl to enforce stricter return code checking, left
a small number of instances behind where return codes were being swallowed
(typically because the function they were being called from was declared as
void). This commit fixes those instances to handle the return codes more
appropriately.

Reviewed-by: Richard Levitte <levitte@openssl.org>

make update

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move internal only ASN.1 functions to asn1_locl.h

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove X509_ATTRIBUTE hack.

The X509_ATTRIBUTE structure includes a hack to tolerate malformed
attributes that encode as the type instead of SET OF type. This form
is never created by OpenSSL and shouldn't be needed any more.

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

free NULL cleanup

This commit handles BIO_ACCEPT_free BIO_CB_FREE BIO_CONNECT_free
BIO_free BIO_free_all BIO_vfree

Reviewed-by: Matt Caswell <matt@openssl.org>

Support key loading from certificate file

Support loading of key and certificate from the same file if
SSL_CONF_FLAG_REQUIRE_PRIVATE is set. This is done by remembering the
filename used for each certificate type and attempting to load a private
key from the file when SSL_CONF_CTX_finish is called.

Update docs.

Reviewed-by: Richard Levitte <levitte@openssl.org>

make depend

Reviewed-by: Richard Levitte <levitte@openssl.org>

make X509_NAME opaque

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix bug in s_client. Previously default verify locations would only be loaded
if CAfile or CApath were also supplied and successfully loaded first.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix HMAC to pass invalid key len test

Reviewed-by: Richard Levitte <levitte@openssl.org>

Add HMAC test for invalid key len

Reviewed-by: Richard Levitte <levitte@openssl.org>

Ensure that both the MD and key have been initialised before attempting to
create an HMAC

Inspired by BoringSSL commit 2fe7f2d0d9a6fcc75b4e594eeec306cc55acd594

Reviewed-by: Richard Levitte <levitte@openssl.org>

Add more HMAC tests

Reviewed-by: Richard Levitte <levitte@openssl.org>

SSL_OP_NETSCAPE_REUSE_CIPHER_CHANGE_BUG was disabled in 0.9.8q and 1.0.0c.
This commit sets the value of SSL_OP_NETSCAPE_REUSE_CIPHER_CHANGE_BUG to
zero.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Deprecate RAND_pseudo_bytes

The justification for RAND_pseudo_bytes is somewhat dubious, and the reality
is that it is frequently being misused. RAND_bytes and RAND_pseudo_bytes in
the default implementation both end up calling ssleay_rand_bytes. Both may
return -1 in an error condition. If there is insufficient entropy then
both will return 0, but RAND_bytes will additionally add an error to the
error queue. They both return 1 on success.
Therefore the fundamental difference between the two is that one will add an
error to the error queue with insufficient entory whilst the other will not.
Frequently there are constructions of this form:

if(RAND_pseudo_bytes(...) <= 1)
goto err;

In the above form insufficient entropy is treated as an error anyway, so
RAND_bytes is probably the better form to use.

This form is also seen:
if(!RAND_pseudo_bytes(...))
goto err;

This is technically not correct at all since a -1 return value is
incorrectly handled - but this form will also treat insufficient entropy as
an error.

Within libssl it is required that you have correctly seeded your entropy
pool and so there seems little benefit in using RAND_pseudo_bytes.
Similarly in libcrypto many operations also require a correctly seeded
entropy pool and so in most interesting cases you would be better off
using RAND_bytes anyway. There is a significant risk of RAND_pseudo_bytes
being incorrectly used in scenarios where security can be compromised by
insufficient entropy.

If you are not using the default implementation, then most engines use the
same function to implement RAND_bytes and RAND_pseudo_bytes in any case.

Given its misuse, limited benefit, and potential to compromise security,
RAND_pseudo_bytes has been deprecated.

Reviewed-by: Richard Levitte <levitte@openssl.org>

RAND_bytes updates

Ensure RAND_bytes return value is checked correctly, and that we no longer
use RAND_pseudo_bytes.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix return checks in GOST engine

Filled in lots of return value checks that were missing the GOST engine, and
added appropriate error handling.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix misc NULL derefs in sureware engine

Fix miscellaneous NULL pointer derefs in the sureware engine.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Add ticket length before buffering DTLS message

In ssl3_send_new_session_ticket the message to be sent is constructed. We
skip adding the length of the session ticket initially, then call
ssl_set_handshake_header, and finally go back and add in the length of the
ticket. Unfortunately, in DTLS, ssl_set_handshake_header also has the side
effect of buffering the message for subsequent retransmission if required.
By adding the ticket length after the call to ssl_set_handshake_header the
message that is buffered is incomplete, causing an invalid message to be
sent on retransmission.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Ensure last_write_sequence is saved in DTLS1.2

In DTLS, immediately prior to epoch change, the write_sequence is supposed
to be stored in s->d1->last_write_sequence. The write_sequence is then reset
back to 00000000. In the event of retransmits of records from the previous
epoch, the last_write_sequence is restored. This commit fixes a bug in
DTLS1.2 where the write_sequence was being reset before last_write_sequence
was saved, and therefore retransmits are sent with incorrect sequence
numbers.

Reviewed-by: Richard Levitte <levitte@openssl.org>

free NULL cleanup

Start ensuring all OpenSSL "free" routines allow NULL, and remove
any if check before calling them.
This gets DH_free, DSA_free, RSA_free

Reviewed-by: Matt Caswell <matt@openssl.org>

update ordinals

Reviewed-by: Richard Levitte <levitte@openssl.org>

Update ordinals

Thanks to the change of mkdef.pl, a few more deprecated functions were
properly defined in util/libeay.num.

Reviewed-by: Matt Caswell <matt@openssl.org>

Teach mkdef.pl to handle multiline declarations.

For the moment, this is specially crafted for DECLARE_DEPRECATED because
that's where we found the problem, but it can easily be expanded to other
types of special delarations when needed.

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix verify algorithm.

Disable loop checking when we retry verification with an alternative path.
This fixes the case where an intermediate CA is explicitly trusted and part
of the untrusted certificate list. By disabling loop checking for this case
the untrusted CA can be replaced by the explicitly trusted case and
verification will succeed.

Reviewed-by: Matt Caswell <matt@openssl.org>

make ASN1_OBJECT opaque

Reviewed-by: Matt Caswell <matt@openssl.org>

Configuration file examples.

Reviewed-by: Matt Caswell <matt@openssl.org>

Make OCSP response verification more flexible.

If a set of certificates is supplied to OCSP_basic_verify use those in
addition to any present in the OCSP response as untrusted CAs when
verifying a certificate chain.

PR#3668

Reviewed-by: Matt Caswell <matt@openssl.org>

make depend

Reviewed-by: Matt Caswell <matt@openssl.org>

Move some EVP internals to evp_int.h

Move EVP internals to evp_int.h, remove -Ievp hack from crypto/Makefile

Reviewed-by: Matt Caswell <matt@openssl.org>

Move some ASN.1 internals to asn1_int.h

Move ASN.1 internals used across multiple directories into new internal
header file asn1_int.h remove crypto/Makefile hack which allowed other
directories to include "asn1_locl.h"

Reviewed-by: Matt Caswell <matt@openssl.org>

free NULL cleanup

Start ensuring all OpenSSL "free" routines allow NULL, and remove
any if check before calling them.
This gets ASN1_OBJECT_free and ASN1_STRING_free.

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix malloc define typo

Fix compilation failure when SCTP is compiled due to incorrect define.

Reported-by: Conrad Kostecki <ck+gentoobugzilla@bl4ckb0x.de>
URL: https://bugs.gentoo.org/543828

RT#3758
Signed-off-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>

Use OPENSSL_malloc rather than malloc/calloc

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix eng_cryptodev to not depend on BN internals.

Reviewed-by: Matt Caswell <matt@openssl.org>

Adjust include path

Thanks to a -I.., the path does work, at least on unix.  However, this
doesn't work so well on VMS.  Correcting the path to not rely on given
-I does work on both.

Reviewed-by: Matt Caswell <matt@openssl.org>

JPAKE Makefile missing 'files' target

Reviewed-by: Matt Caswell <matt@openssl.org>

Remove old style ASN.1 support.

Remove old ASN.1 COMPAT type. This was meant as a temporary measure
so older ASN.1 code (from OpenSSL 0.9.6) still worked. It's a hack
which breaks constification and hopefully nothing uses it now, if
it ever did.

Reviewed-by: Matt Caswell <matt@openssl.org>

return unexpected message when receiving kx with kDHr or kDHd

It was saying that it was an illegal parameter / unsupported cipher

Reviewed-by: Matt Caswell <matt@openssl.org>

Don't send a for ServerKeyExchange for kDHr and kDHd

The certificate already contains the DH parameters in that case.
ssl3_send_server_key_exchange() would fail in that case anyway.

Reviewed-by: Matt Caswell <matt@openssl.org>

Make sure that cert is never NULL

Also removes for it being NULL

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix build.

Remove x_exten.c and x_exten.o from crypto/asn1/Makefile: they've moved now.

Reviewed-by: Matt Caswell <matt@openssl.org>

make X509_EXTENSION opaque

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix SSL_clear unused return

Fix missing return value check in dtls1_listen when calling SSL_clear().

Reviewed-by: Richard Levitte <levitte@openssl.org>

ssl3_set_handshake_header returns

Change ssl_set_handshake_header from return void to returning int, and
handle error return code appropriately.

Reviewed-by: Richard Levitte <levitte@openssl.org>

apps return value checks

Ensure that all libssl functions called from within the apps have their
return values checked where appropriate.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix missing return value checks

Ensure that all functions have their return values checked where
appropriate. This covers all functions defined and called from within
libssl.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Check libssl function returns

Mark most functions returning a result defined in any libssl header file
with __owur to warn if they are used without checking the return value.
Use -DUNUSED_RETURN compiler flag with gcc to activate these warnings.
Some functions returning a result are skipped if it is common and valid to
use these functions without checking the return value.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Add -DDEBUG_UNUSED to --strict-warnings

In order to receive warnings on unused function return values the flag
-DDEBUG_UNUSED must be passed to the compiler. This change adds that for the
--strict-warnings Configure option.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove PREFIX, as it's not used any more.

Reviewed-by: Matt Caswell <matt@openssl.org>

Actually remove TABLE from version control

Follow up on the earlier "Do not keep TABLE in version control".
Actually removing TABLE from version control was forgotten.

Reviewed-by: Matt Caswell <matt@openssl.org>

Don't check curves that haven't been sent

Don't check that the curve appears in the list of acceptable curves for the
peer, if they didn't send us such a list (RFC 4492 does not require that the
extension be sent).

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Remove deleted functions, update ordinals.

Reviewed-by: Matt Caswell <matt@openssl.org>

Remove {i2d,d2i}_ASN1_BOOLEAN

Remove {i2d,d2i}_ASN1_BOOLEAN.

Rewrite single occurrence of d2i_ASN1_BOOLEAN in asn1_parse2

Reviewed-by: Rich Salz <rsalz@openssl.org>

Remove old ASN.1 code.

Remove old M_ASN1_ macros and replace any occurences with the corresponding
function.

Remove d2i_ASN1_bytes, d2i_ASN1_SET, i2d_ASN1_SET: no longer used internally.

Reviewed-by: Rich Salz <rsalz@openssl.org>

sha/asm/sha256-armv4.pl: adapt for use in Linux kernel context.

In cooperation with Ard Biesheuvel (Linaro) and Sami Tolvanen (Google).

Reviewed-by: Richard Levitte <levitte@openssl.org>

Refer to $table{$target} rather than $table{$t}.

Using $t is an artifact from the earlier changes in Configure and was
unfortunately forgotten as is.

Reviewed-by: Stephen Henson <steve@openssl.org>

Add AES unwrap test with invalid key.

This tests the unwrap algorithm with an invalid key. The result should
be rejected without returning any plaintext.

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Fix memory leak.

Reviewed-by: Emilia Käsper <emilia@openssl.org>

CRYPTO_128_unwrap(): Fix refactoring damage

crypto/modes/wrap128.c was heavily refactored to support AES Key Wrap
with Padding, and four bugs were introduced into CRYPTO_128_unwrap() at
that time:

- crypto_128_unwrap_raw()'s return value ('ret') is checked incorrectly,
  and the function immediately returns 'ret' in (almost) all cases.
  This makes the IV checking code later in the function unreachable, but
  callers think the IV check succeeded since CRYPTO_128_unwrap()'s
  return value is non-zero.

  FIX: Return 0 (error) if crypto_128_unwrap_raw() returned 0 (error).

- crypto_128_unwrap_raw() writes the IV to the 'got_iv' buffer, not to
  the first 8 bytes of the output buffer ('out') as the IV checking code
  expects.  This makes the IV check fail.

  FIX: Compare 'iv' to 'got_iv', not 'out'.

- The data written to the output buffer ('out') is "cleansed" if the IV
  check fails, but the code passes OPENSSL_cleanse() the input buffer
  length ('inlen') instead of the number of bytes that
  crypto_128_unwrap_raw() wrote to the output buffer ('ret').  This
  means that OPENSSL_cleanse() could potentially write past the end of
  'out'.

  FIX: Change 'inlen' to 'ret' in the OPENSSL_cleanse() call.

- CRYPTO_128_unwrap() is returning the length of the input buffer
  ('inlen') instead of the number of bytes written to the output buffer
  ('ret').  This could cause the caller to read past the end of 'out'.

  FIX: Return 'ret' instead of 'inlen' at the end of the function.

PR#3749

Reviewed-by: Stephen Henson <steve@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

wrap128.c: Fix Doxygen comments

Reviewed-by: Stephen Henson <steve@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

Add DTLS tests to make test

Updated test/testssl script to include the new DTLS capability in ssltest.

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Add DTLS support to ssltest

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Add DTLS to SSL_get_version

Reviewed-by: Emilia Käsper <emilia@openssl.org>

If the target is an old style debug- target, it will not have debugging [cl]flags

Reviewed-by: Stephen Henson <steve@openssl.org>

Fix a failure to NULL a pointer freed on error.

Reported by the LibreSSL project as a follow on to CVE-2015-0209

Reviewed-by: Richard Levitte <levitte@openssl.org>

Update NEWS

Resync NEWS with the latest version from 1.0.2

Reviewed-by: Richard Levitte <levitte@openssl.org>

Update CHANGES

Resync CHANGES with the latest version from 1.0.2.

Reviewed-by: Richard Levitte <levitte@openssl.org>

PKCS#7: avoid NULL pointer dereferences with missing content

In PKCS#7, the ASN.1 content component is optional.
This typically applies to inner content (detached signatures),
however we must also handle unexpected missing outer content
correctly.

This patch only addresses functions reachable from parsing,
decryption and verification, and functions otherwise associated
with reading potentially untrusted data.

Correcting all low-level API calls requires further work.

CVE-2015-0289

Thanks to Michal Zalewski (Google) for reporting this issue.

Reviewed-by: Steve Henson <steve@openssl.org>

Fix ASN1_TYPE_cmp

Fix segmentation violation when ASN1_TYPE_cmp is passed a boolean type. This
can be triggered during certificate verification so could be a DoS attack
against a client or a server enabling client authentication.

CVE-2015-0286

Reviewed-by: Richard Levitte <levitte@openssl.org>