CHANGES

     cache-timing and potential related attacks.

     BN_mod_exp_mont_consttime() is the new exponentiation implementation,
     and this is automatically used by BN_mod_exp_mont() if the new flag
     BN_FLG_EXP_CONSTTIME is set for the exponent.  RSA, DSA, and DH
     will use this BN flag for private exponents unless the flag
     RSA_FLAG_NO_EXP_CONSTTIME, DSA_FLAG_NO_EXP_CONSTTIME, or
     DH_FLAG_NO_EXP_CONSTTIME, respectively, is set.

     [Matthew D Wood (Intel Corp), with some changes by Bodo Moeller]

  *) Change the client implementation for SSLv23_method() and
     SSLv23_client_method() so that is uses the SSL 3.0/TLS 1.0
     Client Hello message format if the SSL_OP_NO_SSLv2 option is set.
     (Previously, the SSL 2.0 backwards compatible Client Hello
     message format would be used even with SSL_OP_NO_SSLv2.)
     [Bodo Moeller]

  *) Add support for smime-type MIME parameter in S/MIME messages which some
     clients need.
     [Steve Henson]

  *) New function BN_MONT_CTX_set_locked() to set montgomery parameters in
     a threadsafe manner. Modify rsa code to use new function and add calls
     to dsa and dh code (which had race conditions before).
     [Steve Henson]

  *) Include the fixed error library code in the C error file definitions
     instead of fixing them up at runtime. This keeps the error code
     structures constant.
     [Steve Henson]

 Changes between 0.9.7f and 0.9.7g  [11 Apr 2005]

  [NB: OpenSSL 0.9.7h and later 0.9.7 patch levels were released after
  OpenSSL 0.9.8.]

  *) Fixes for newer kerberos headers. NB: the casts are needed because
     the 'length' field is signed on one version and unsigned on another
     with no (?) obvious way to tell the difference, without these VC++
     complains. Also the "definition" of FAR (blank) is no longer included
     nor is the error ENOMEM. KRB5_PRIVATE has to be set to 1 to pick up
     some needed definitions.
     [Steve Henson]

  *) Undo Cygwin change.
     [Ulf Möller]

  *) Added support for proxy certificates according to RFC 3820.
     Because they may be a security thread to unaware applications,
     they must be explicitely allowed in run-time.  See
     docs/HOWTO/proxy_certificates.txt for further information.
     [Richard Levitte]

 Changes between 0.9.7e and 0.9.7f  [22 Mar 2005]

  *) Use (SSL_RANDOM_VALUE - 4) bytes of pseudo random data when generating
     server and client random values. Previously
     (SSL_RANDOM_VALUE - sizeof(time_t)) would be used which would result in
     less random data when sizeof(time_t) > 4 (some 64 bit platforms).

     This change has negligible security impact because:

     1. Server and client random values still have 24 bytes of pseudo random
        data.

     2. Server and client random values are sent in the clear in the initial
        handshake.

     3. The master secret is derived using the premaster secret (48 bytes in
        size for static RSA ciphersuites) as well as client server and random
        values.

     The OpenSSL team would like to thank the UK NISCC for bringing this issue
     to our attention. 

     [Stephen Henson, reported by UK NISCC]

  *) Use Windows randomness collection on Cygwin.
     [Ulf Möller]

  *) Fix hang in EGD/PRNGD query when communication socket is closed
     prematurely by EGD/PRNGD.
     [Darren Tucker <dtucker@zip.com.au> via Lutz Jänicke, resolves #1014]

  *) Prompt for pass phrases when appropriate for PKCS12 input format.
     [Steve Henson]

  *) Back-port of selected performance improvements from development
     branch, as well as improved support for PowerPC platforms.
     [Andy Polyakov]

  *) Add lots of checks for memory allocation failure, error codes to indicate
     failure and freeing up memory if a failure occurs.
     [Nauticus Networks SSL Team <openssl@nauticusnet.com>, Steve Henson]

  *) Add new -passin argument to dgst.
     [Steve Henson]

  *) Perform some character comparisons of different types in X509_NAME_cmp:
     this is needed for some certificates that reencode DNs into UTF8Strings
     (in violation of RFC3280) and can't or wont issue name rollover
     certificates.
     [Steve Henson]

  *) Make an explicit check during certificate validation to see that
     the CA setting in each certificate on the chain is correct.  As a
     side effect always do the following basic checks on extensions,
     not just when there's an associated purpose to the check:

      - if there is an unhandled critical extension (unless the user
        has chosen to ignore this fault)
      - if the path length has been exceeded (if one is set at all)
      - that certain extensions fit the associated purpose (if one has
        been given)
     [Richard Levitte]

 Changes between 0.9.7d and 0.9.7e  [25 Oct 2004]

  *) Avoid a race condition when CRLs are checked in a multi threaded 
     environment. This would happen due to the reordering of the revoked
     entries during signature checking and serial number lookup. Now the
     encoding is cached and the serial number sort performed under a lock.
     Add new STACK function sk_is_sorted().
     [Steve Henson]

  *) Add Delta CRL to the extension code.
     [Steve Henson]

  *) Various fixes to s3_pkt.c so alerts are sent properly.
     [David Holmes <d.holmes@f5.com>]

  *) Reduce the chances of duplicate issuer name and serial numbers (in
     violation of RFC3280) using the OpenSSL certificate creation utilities.
     This is done by creating a random 64 bit value for the initial serial
     number when a serial number file is created or when a self signed
     certificate is created using 'openssl req -x509'. The initial serial
     number file is created using 'openssl x509 -next_serial' in CA.pl
     rather than being initialized to 1.
     [Steve Henson]

 Changes between 0.9.7c and 0.9.7d  [17 Mar 2004]

  *) Fix null-pointer assignment in do_change_cipher_spec() revealed           
     by using the Codenomicon TLS Test Tool (CVE-2004-0079)                    
     [Joe Orton, Steve Henson]   

  *) Fix flaw in SSL/TLS handshaking when using Kerberos ciphersuites
     (CVE-2004-0112)
     [Joe Orton, Steve Henson]   

  *) Make it possible to have multiple active certificates with the same
     subject in the CA index file.  This is done only if the keyword
     'unique_subject' is set to 'no' in the main CA section (default
     if 'CA_default') of the configuration file.  The value is saved
     with the database itself in a separate index attribute file,
     named like the index file with '.attr' appended to the name.
     [Richard Levitte]

  *) X509 verify fixes. Disable broken certificate workarounds when 
     X509_V_FLAGS_X509_STRICT is set. Check CRL issuer has cRLSign set if
     keyUsage extension present. Don't accept CRLs with unhandled critical
     extensions: since verify currently doesn't process CRL extensions this
     rejects a CRL with *any* critical extensions. Add new verify error codes
     for these cases.
     [Steve Henson]

  *) When creating an OCSP nonce use an OCTET STRING inside the extnValue.
     A clarification of RFC2560 will require the use of OCTET STRINGs and 
     some implementations cannot handle the current raw format. Since OpenSSL
     copies and compares OCSP nonces as opaque blobs without any attempt at
     parsing them this should not create any compatibility issues.
     [Steve Henson]

  *) New md flag EVP_MD_CTX_FLAG_REUSE this allows md_data to be reused when
     calling EVP_MD_CTX_copy_ex() to avoid calling OPENSSL_malloc(). Without
     this HMAC (and other) operations are several times slower than OpenSSL
     < 0.9.7.
     [Steve Henson]

  *) Print out GeneralizedTime and UTCTime in ASN1_STRING_print_ex().
     [Peter Sylvester <Peter.Sylvester@EdelWeb.fr>]

  *) Use the correct content when signing type "other".
     [Steve Henson]

 Changes between 0.9.7b and 0.9.7c  [30 Sep 2003]

  *) Fix various bugs revealed by running the NISCC test suite:

     Stop out of bounds reads in the ASN1 code when presented with
     invalid tags (CVE-2003-0543 and CVE-2003-0544).
     
     Free up ASN1_TYPE correctly if ANY type is invalid (CVE-2003-0545).

     If verify callback ignores invalid public key errors don't try to check
     certificate signature with the NULL public key.

     [Steve Henson]

  *) New -ignore_err option in ocsp application to stop the server
     exiting on the first error in a request.
     [Steve Henson]

  *) In ssl3_accept() (ssl/s3_srvr.c) only accept a client certificate
     if the server requested one: as stated in TLS 1.0 and SSL 3.0
     specifications.
     [Steve Henson]

  *) In ssl3_get_client_hello() (ssl/s3_srvr.c), tolerate additional
     extra data after the compression methods not only for TLS 1.0
     but also for SSL 3.0 (as required by the specification).
     [Bodo Moeller; problem pointed out by Matthias Loepfe]

  *) Change X509_certificate_type() to mark the key as exported/exportable
     when it's 512 *bits* long, not 512 bytes.
     [Richard Levitte]

  *) Change AES_cbc_encrypt() so it outputs exact multiple of
     blocks during encryption.
     [Richard Levitte]

  *) Various fixes to base64 BIO and non blocking I/O. On write 
     flushes were not handled properly if the BIO retried. On read
     data was not being buffered properly and had various logic bugs.
     This also affects blocking I/O when the data being decoded is a
     certain size.
     [Steve Henson]

  *) Various S/MIME bugfixes and compatibility changes:
     output correct application/pkcs7 MIME type if
     PKCS7_NOOLDMIMETYPE is set. Tolerate some broken signatures.
     Output CR+LF for EOL if PKCS7_CRLFEOL is set (this makes opening
     of files as .eml work). Correctly handle very long lines in MIME
     parser.
     [Steve Henson]

 Changes between 0.9.7a and 0.9.7b  [10 Apr 2003]

  *) Countermeasure against the Klima-Pokorny-Rosa extension of
     Bleichbacher's attack on PKCS #1 v1.5 padding: treat
     a protocol version number mismatch like a decryption error
     in ssl3_get_client_key_exchange (ssl/s3_srvr.c).
     [Bodo Moeller]

  *) Turn on RSA blinding by default in the default implementation
     to avoid a timing attack. Applications that don't want it can call
     RSA_blinding_off() or use the new flag RSA_FLAG_NO_BLINDING.
     They would be ill-advised to do so in most cases.
     [Ben Laurie, Steve Henson, Geoff Thorpe, Bodo Moeller]

  *) Change RSA blinding code so that it works when the PRNG is not
     seeded (in this case, the secret RSA exponent is abused as
     an unpredictable seed -- if it is not unpredictable, there
     is no point in blinding anyway).  Make RSA blinding thread-safe
     by remembering the creator's thread ID in rsa->blinding and
     having all other threads use local one-time blinding factors
     (this requires more computation than sharing rsa->blinding, but
     avoids excessive locking; and if an RSA object is not shared
     between threads, blinding will still be very fast).
     [Bodo Moeller]

  *) Fixed a typo bug that would cause ENGINE_set_default() to set an
     ENGINE as defaults for all supported algorithms irrespective of
     the 'flags' parameter. 'flags' is now honoured, so applications
     should make sure they are passing it correctly.
     [Geoff Thorpe]

  *) Target "mingw" now allows native Windows code to be generated in
     the Cygwin environment as well as with the MinGW compiler.
     [Ulf Moeller] 

 Changes between 0.9.7 and 0.9.7a  [19 Feb 2003]

  *) In ssl3_get_record (ssl/s3_pkt.c), minimize information leaked
     via timing by performing a MAC computation even if incorrect
     block cipher padding has been found.  This is a countermeasure
     against active attacks where the attacker has to distinguish
     between bad padding and a MAC verification error. (CVE-2003-0078)

     [Bodo Moeller; problem pointed out by Brice Canvel (EPFL),
     Alain Hiltgen (UBS), Serge Vaudenay (EPFL), and
     Martin Vuagnoux (EPFL, Ilion)]

  *) Make the no-err option work as intended.  The intention with no-err
     is not to have the whole error stack handling routines removed from
     libcrypto, it's only intended to remove all the function name and
     reason texts, thereby removing some of the footprint that may not
     be interesting if those errors aren't displayed anyway.

     NOTE: it's still possible for any application or module to have it's
     own set of error texts inserted.  The routines are there, just not
     used by default when no-err is given.
     [Richard Levitte]

  *) Add support for FreeBSD on IA64.
     [dirk.meyer@dinoex.sub.org via Richard Levitte, resolves #454]

  *) Adjust DES_cbc_cksum() so it returns the same value as the MIT
     Kerberos function mit_des_cbc_cksum().  Before this change,
     the value returned by DES_cbc_cksum() was like the one from
     mit_des_cbc_cksum(), except the bytes were swapped.
     [Kevin Greaney <Kevin.Greaney@hp.com> and Richard Levitte]

  *) Allow an application to disable the automatic SSL chain building.
     Before this a rather primitive chain build was always performed in
     ssl3_output_cert_chain(): an application had no way to send the 
     correct chain if the automatic operation produced an incorrect result.

     Now the chain builder is disabled if either:

     1. Extra certificates are added via SSL_CTX_add_extra_chain_cert().

     2. The mode flag SSL_MODE_NO_AUTO_CHAIN is set.

     The reasoning behind this is that an application would not want the
     auto chain building to take place if extra chain certificates are
     present and it might also want a means of sending no additional
     certificates (for example the chain has two certificates and the
     root is omitted).
     [Steve Henson]

  *) Add the possibility to build without the ENGINE framework.
     [Steven Reddie <smr@essemer.com.au> via Richard Levitte]

  *) Under Win32 gmtime() can return NULL: check return value in
     OPENSSL_gmtime(). Add error code for case where gmtime() fails.
     [Steve Henson]

  *) DSA routines: under certain error conditions uninitialized BN objects
     could be freed. Solution: make sure initialization is performed early
     enough. (Reported and fix supplied by Ivan D Nestlerode <nestler@MIT.EDU>,
     Nils Larsch <nla@trustcenter.de> via PR#459)
     [Lutz Jaenicke]

  *) Another fix for SSLv2 session ID handling: the session ID was incorrectly
     checked on reconnect on the client side, therefore session resumption
     could still fail with a "ssl session id is different" error. This
     behaviour is masked when SSL_OP_ALL is used due to
     SSL_OP_MICROSOFT_SESS_ID_BUG being set.
     Behaviour observed by Crispin Flowerday <crispin@flowerday.cx> as
     followup to PR #377.
     [Lutz Jaenicke]

  *) IA-32 assembler support enhancements: unified ELF targets, support
     for SCO/Caldera platforms, fix for Cygwin shared build.
     [Andy Polyakov]

  *) Add support for FreeBSD on sparc64.  As a consequence, support for
     FreeBSD on non-x86 processors is separate from x86 processors on
     the config script, much like the NetBSD support.
     [Richard Levitte & Kris Kennaway <kris@obsecurity.org>]

 Changes between 0.9.6h and 0.9.7  [31 Dec 2002]

  [NB: OpenSSL 0.9.6i and later 0.9.6 patch levels were released after
  OpenSSL 0.9.7.]

  *) Fix session ID handling in SSLv2 client code: the SERVER FINISHED
     code (06) was taken as the first octet of the session ID and the last
     octet was ignored consequently. As a result SSLv2 client side session
     caching could not have worked due to the session ID mismatch between
     client and server.
     Behaviour observed by Crispin Flowerday <crispin@flowerday.cx> as
     PR #377.
     [Lutz Jaenicke]

  *) Change the declaration of needed Kerberos libraries to use EX_LIBS
     instead of the special (and badly supported) LIBKRB5.  LIBKRB5 is
     removed entirely.
     [Richard Levitte]

  *) The hw_ncipher.c engine requires dynamic locks.  Unfortunately, it
     seems that in spite of existing for more than a year, many application
     author have done nothing to provide the necessary callbacks, which
     means that this particular engine will not work properly anywhere.
     This is a very unfortunate situation which forces us, in the name
     of usability, to give the hw_ncipher.c a static lock, which is part
     of libcrypto.
     NOTE: This is for the 0.9.7 series ONLY.  This hack will never
     appear in 0.9.8 or later.  We EXPECT application authors to have
     dealt properly with this when 0.9.8 is released (unless we actually
     make such changes in the libcrypto locking code that changes will
     have to be made anyway).
     [Richard Levitte]

  *) In asn1_d2i_read_bio() repeatedly call BIO_read() until all content
     octets have been read, EOF or an error occurs. Without this change
     some truncated ASN1 structures will not produce an error.
     [Steve Henson]

  *) Disable Heimdal support, since it hasn't been fully implemented.
     Still give the possibility to force the use of Heimdal, but with
     warnings and a request that patches get sent to openssl-dev.
     [Richard Levitte]

  *) Add the VC-CE target, introduce the WINCE sysname, and add
     INSTALL.WCE and appropriate conditionals to make it build.
     [Steven Reddie <smr@essemer.com.au> via Richard Levitte]

  *) Change the DLL names for Cygwin to cygcrypto-x.y.z.dll and
     cygssl-x.y.z.dll, where x, y and z are the major, minor and
     edit numbers of the version.
     [Corinna Vinschen <vinschen@redhat.com> and Richard Levitte]

  *) Introduce safe string copy and catenation functions
     (BUF_strlcpy() and BUF_strlcat()).
     [Ben Laurie (CHATS) and Richard Levitte]

  *) Avoid using fixed-size buffers for one-line DNs.
     [Ben Laurie (CHATS)]

  *) Add BUF_MEM_grow_clean() to avoid information leakage when
     resizing buffers containing secrets, and use where appropriate.
     [Ben Laurie (CHATS)]

  *) Avoid using fixed size buffers for configuration file location.
     [Ben Laurie (CHATS)]

  *) Avoid filename truncation for various CA files.
     [Ben Laurie (CHATS)]

  *) Use sizeof in preference to magic numbers.
     [Ben Laurie (CHATS)]

  *) Avoid filename truncation in cert requests.
     [Ben Laurie (CHATS)]

  *) Add assertions to check for (supposedly impossible) buffer
     overflows.
     [Ben Laurie (CHATS)]

  *) Don't cache truncated DNS entries in the local cache (this could
     potentially lead to a spoofing attack).
     [Ben Laurie (CHATS)]

  *) Fix various buffers to be large enough for hex/decimal
     representations in a platform independent manner.
     [Ben Laurie (CHATS)]

  *) Add CRYPTO_realloc_clean() to avoid information leakage when
     resizing buffers containing secrets, and use where appropriate.
     [Ben Laurie (CHATS)]

  *) Add BIO_indent() to avoid much slightly worrying code to do
     indents.
     [Ben Laurie (CHATS)]

  *) Convert sprintf()/BIO_puts() to BIO_printf().
     [Ben Laurie (CHATS)]

  *) buffer_gets() could terminate with the buffer only half
     full. Fixed.
     [Ben Laurie (CHATS)]

  *) Add assertions to prevent user-supplied crypto functions from
     overflowing internal buffers by having large block sizes, etc.
     [Ben Laurie (CHATS)]

  *) New OPENSSL_assert() macro (similar to assert(), but enabled
     unconditionally).
     [Ben Laurie (CHATS)]

  *) Eliminate unused copy of key in RC4.
     [Ben Laurie (CHATS)]

  *) Eliminate unused and incorrectly sized buffers for IV in pem.h.
     [Ben Laurie (CHATS)]

  *) Fix off-by-one error in EGD path.
     [Ben Laurie (CHATS)]

  *) If RANDFILE path is too long, ignore instead of truncating.
     [Ben Laurie (CHATS)]

  *) Eliminate unused and incorrectly sized X.509 structure
     CBCParameter.
     [Ben Laurie (CHATS)]

  *) Eliminate unused and dangerous function knumber().
     [Ben Laurie (CHATS)]

  *) Eliminate unused and dangerous structure, KSSL_ERR.
     [Ben Laurie (CHATS)]

  *) Protect against overlong session ID context length in an encoded
     session object. Since these are local, this does not appear to be
     exploitable.
     [Ben Laurie (CHATS)]

  *) Change from security patch (see 0.9.6e below) that did not affect
     the 0.9.6 release series:

     Remote buffer overflow in SSL3 protocol - an attacker could
     supply an oversized master key in Kerberos-enabled versions.
     (CVE-2002-0657)
     [Ben Laurie (CHATS)]

  *) Change the SSL kerb5 codes to match RFC 2712.
     [Richard Levitte]

  *) Make -nameopt work fully for req and add -reqopt switch.
     [Michael Bell <michael.bell@rz.hu-berlin.de>, Steve Henson]

  *) The "block size" for block ciphers in CFB and OFB mode should be 1.
     [Steve Henson, reported by Yngve Nysaeter Pettersen <yngve@opera.com>]

  *) Make sure tests can be performed even if the corresponding algorithms
     have been removed entirely.  This was also the last step to make
     OpenSSL compilable with DJGPP under all reasonable conditions.
     [Richard Levitte, Doug Kaufman <dkaufman@rahul.net>]

  *) Add cipher selection rules COMPLEMENTOFALL and COMPLEMENTOFDEFAULT
     to allow version independent disabling of normally unselected ciphers,
     which may be activated as a side-effect of selecting a single cipher.

     (E.g., cipher list string "RSA" enables ciphersuites that are left
     out of "ALL" because they do not provide symmetric encryption.
     "RSA:!COMPLEMEMENTOFALL" avoids these unsafe ciphersuites.)
     [Lutz Jaenicke, Bodo Moeller]

  *) Add appropriate support for separate platform-dependent build
     directories.  The recommended way to make a platform-dependent
     build directory is the following (tested on Linux), maybe with
     some local tweaks:

	# Place yourself outside of the OpenSSL source tree.  In
	# this example, the environment variable OPENSSL_SOURCE
	# is assumed to contain the absolute OpenSSL source directory.
	mkdir -p objtree/"`uname -s`-`uname -r`-`uname -m`"
	cd objtree/"`uname -s`-`uname -r`-`uname -m`"
	(cd $OPENSSL_SOURCE; find . -type f) | while read F; do
		mkdir -p `dirname $F`
		ln -s $OPENSSL_SOURCE/$F $F
	done

     To be absolutely sure not to disturb the source tree, a "make clean"
     is a good thing.  If it isn't successful, don't worry about it,
     it probably means the source directory is very clean.
     [Richard Levitte]

  *) Make sure any ENGINE control commands make local copies of string
     pointers passed to them whenever necessary. Otherwise it is possible
     the caller may have overwritten (or deallocated) the original string
     data when a later ENGINE operation tries to use the stored values.
     [Götz Babin-Ebell <babinebell@trustcenter.de>]

  *) Improve diagnostics in file reading and command-line digests.
     [Ben Laurie aided and abetted by Solar Designer <solar@openwall.com>]

  *) Add AES modes CFB and OFB to the object database.  Correct an
     error in AES-CFB decryption.
     [Richard Levitte]

  *) Remove most calls to EVP_CIPHER_CTX_cleanup() in evp_enc.c, this 
     allows existing EVP_CIPHER_CTX structures to be reused after
     calling EVP_*Final(). This behaviour is used by encryption
     BIOs and some applications. This has the side effect that
     applications must explicitly clean up cipher contexts with
     EVP_CIPHER_CTX_cleanup() or they will leak memory.
     [Steve Henson]

  *) Check the values of dna and dnb in bn_mul_recursive before calling
     bn_mul_comba (a non zero value means the a or b arrays do not contain
     n2 elements) and fallback to bn_mul_normal if either is not zero.
     [Steve Henson]

  *) Fix escaping of non-ASCII characters when using the -subj option
     of the "openssl req" command line tool. (Robert Joop <joop@fokus.gmd.de>)
     [Lutz Jaenicke]

  *) Make object definitions compliant to LDAP (RFC2256): SN is the short
     form for "surname", serialNumber has no short form.
     Use "mail" as the short name for "rfc822Mailbox" according to RFC2798;
     therefore remove "mail" short name for "internet 7".
     The OID for unique identifiers in X509 certificates is
     x500UniqueIdentifier, not uniqueIdentifier.
     Some more OID additions. (Michael Bell <michael.bell@rz.hu-berlin.de>)
     [Lutz Jaenicke]

  *) Add an "init" command to the ENGINE config module and auto initialize
     ENGINEs. Without any "init" command the ENGINE will be initialized 
     after all ctrl commands have been executed on it. If init=1 the 
     ENGINE is initailized at that point (ctrls before that point are run
     on the uninitialized ENGINE and after on the initialized one). If
     init=0 then the ENGINE will not be iniatialized at all.
     [Steve Henson]

  *) Fix the 'app_verify_callback' interface so that the user-defined
     argument is actually passed to the callback: In the
     SSL_CTX_set_cert_verify_callback() prototype, the callback
     declaration has been changed from
          int (*cb)()
     into
          int (*cb)(X509_STORE_CTX *,void *);
     in ssl_verify_cert_chain (ssl/ssl_cert.c), the call
          i=s->ctx->app_verify_callback(&ctx)
     has been changed into
          i=s->ctx->app_verify_callback(&ctx, s->ctx->app_verify_arg).

     To update applications using SSL_CTX_set_cert_verify_callback(),
     a dummy argument can be added to their callback functions.
     [D. K. Smetters <smetters@parc.xerox.com>]

  *) Added the '4758cca' ENGINE to support IBM 4758 cards.
     [Maurice Gittens <maurice@gittens.nl>, touchups by Geoff Thorpe]

  *) Add and OPENSSL_LOAD_CONF define which will cause
     OpenSSL_add_all_algorithms() to load the openssl.cnf config file.
     This allows older applications to transparently support certain
     OpenSSL features: such as crypto acceleration and dynamic ENGINE loading.
     Two new functions OPENSSL_add_all_algorithms_noconf() which will never
     load the config file and OPENSSL_add_all_algorithms_conf() which will
     always load it have also been added.
     [Steve Henson]

  *) Add the OFB, CFB and CTR (all with 128 bit feedback) to AES.
     Adjust NIDs and EVP layer.
     [Stephen Sprunk <stephen@sprunk.org> and Richard Levitte]

  *) Config modules support in openssl utility.

     Most commands now load modules from the config file,
     though in a few (such as version) this isn't done 
     because it couldn't be used for anything.

     In the case of ca and req the config file used is
     the same as the utility itself: that is the -config
     command line option can be used to specify an
     alternative file.
     [Steve Henson]

  *) Move default behaviour from OPENSSL_config(). If appname is NULL
     use "openssl_conf" if filename is NULL use default openssl config file.
     [Steve Henson]

  *) Add an argument to OPENSSL_config() to allow the use of an alternative
     config section name. Add a new flag to tolerate a missing config file
     and move code to CONF_modules_load_file().
     [Steve Henson]

  *) Support for crypto accelerator cards from Accelerated Encryption
     Processing, www.aep.ie.  (Use engine 'aep')
     The support was copied from 0.9.6c [engine] and adapted/corrected
     to work with the new engine framework.
     [AEP Inc. and Richard Levitte]

  *) Support for SureWare crypto accelerator cards from Baltimore
     Technologies.  (Use engine 'sureware')
     The support was copied from 0.9.6c [engine] and adapted
     to work with the new engine framework.
     [Richard Levitte]

  *) Have the CHIL engine fork-safe (as defined by nCipher) and actually
     make the newer ENGINE framework commands for the CHIL engine work.
     [Toomas Kiisk <vix@cyber.ee> and Richard Levitte]

  *) Make it possible to produce shared libraries on ReliantUNIX.
     [Robert Dahlem <Robert.Dahlem@ffm2.siemens.de> via Richard Levitte]

  *) Add the configuration target debug-linux-ppro.
     Make 'openssl rsa' use the general key loading routines
     implemented in apps.c, and make those routines able to
     handle the key format FORMAT_NETSCAPE and the variant
     FORMAT_IISSGC.
     [Toomas Kiisk <vix@cyber.ee> via Richard Levitte]

 *) Fix a crashbug and a logic bug in hwcrhk_load_pubkey().
     [Toomas Kiisk <vix@cyber.ee> via Richard Levitte]

  *) Add -keyform to rsautl, and document -engine.
     [Richard Levitte, inspired by Toomas Kiisk <vix@cyber.ee>]

  *) Change BIO_new_file (crypto/bio/bss_file.c) to use new
     BIO_R_NO_SUCH_FILE error code rather than the generic
     ERR_R_SYS_LIB error code if fopen() fails with ENOENT.
     [Ben Laurie]

  *) Add new functions
          ERR_peek_last_error
          ERR_peek_last_error_line
          ERR_peek_last_error_line_data.
     These are similar to
          ERR_peek_error
          ERR_peek_error_line
          ERR_peek_error_line_data,
     but report on the latest error recorded rather than the first one
     still in the error queue.
     [Ben Laurie, Bodo Moeller]
        
  *) default_algorithms option in ENGINE config module. This allows things
     like:
     default_algorithms = ALL
     default_algorithms = RSA, DSA, RAND, CIPHERS, DIGESTS
     [Steve Henson]

  *) Preliminary ENGINE config module.
     [Steve Henson]

  *) New experimental application configuration code.
     [Steve Henson]

  *) Change the AES code to follow the same name structure as all other
     symmetric ciphers, and behave the same way.  Move everything to
     the directory crypto/aes, thereby obsoleting crypto/rijndael.
     [Stephen Sprunk <stephen@sprunk.org> and Richard Levitte]

  *) SECURITY: remove unsafe setjmp/signal interaction from ui_openssl.c.
     [Ben Laurie and Theo de Raadt]

  *) Add option to output public keys in req command.
     [Massimiliano Pala madwolf@openca.org]

  *) Use wNAFs in EC_POINTs_mul() for improved efficiency
     (up to about 10% better than before for P-192 and P-224).
     [Bodo Moeller]

  *) New functions/macros

          SSL_CTX_set_msg_callback(ctx, cb)
          SSL_CTX_set_msg_callback_arg(ctx, arg)
          SSL_set_msg_callback(ssl, cb)
          SSL_set_msg_callback_arg(ssl, arg)

     to request calling a callback function

          void cb(int write_p, int version, int content_type,
                  const void *buf, size_t len, SSL *ssl, void *arg)

     whenever a protocol message has been completely received
     (write_p == 0) or sent (write_p == 1).  Here 'version' is the
     protocol version  according to which the SSL library interprets
     the current protocol message (SSL2_VERSION, SSL3_VERSION, or
     TLS1_VERSION).  'content_type' is 0 in the case of SSL 2.0, or
     the content type as defined in the SSL 3.0/TLS 1.0 protocol
     specification (change_cipher_spec(20), alert(21), handshake(22)).
     'buf' and 'len' point to the actual message, 'ssl' to the
     SSL object, and 'arg' is the application-defined value set by
     SSL[_CTX]_set_msg_callback_arg().

     'openssl s_client' and 'openssl s_server' have new '-msg' options
     to enable a callback that displays all protocol messages.
     [Bodo Moeller]

  *) Change the shared library support so shared libraries are built as
     soon as the corresponding static library is finished, and thereby get
     openssl and the test programs linked against the shared library.
     This still only happens when the keyword "shard" has been given to
     the configuration scripts.

     NOTE: shared library support is still an experimental thing, and
     backward binary compatibility is still not guaranteed.
     ["Maciej W. Rozycki" <macro@ds2.pg.gda.pl> and Richard Levitte]

  *) Add support for Subject Information Access extension.
     [Peter Sylvester <Peter.Sylvester@EdelWeb.fr>]

  *) Make BUF_MEM_grow() behaviour more consistent: Initialise to zero
     additional bytes when new memory had to be allocated, not just
     when reusing an existing buffer.
     [Bodo Moeller]

  *) New command line and configuration option 'utf8' for the req command.
     This allows field values to be specified as UTF8 strings.
     [Steve Henson]

  *) Add -multi and -mr options to "openssl speed" - giving multiple parallel
     runs for the former and machine-readable output for the latter.
     [Ben Laurie]

  *) Add '-noemailDN' option to 'openssl ca'.  This prevents inclusion
     of the e-mail address in the DN (i.e., it will go into a certificate
     extension only).  The new configuration file option 'email_in_dn = no'
     has the same effect.
     [Massimiliano Pala madwolf@openca.org]

  *) Change all functions with names starting with des_ to be starting
     with DES_ instead.  Add wrappers that are compatible with libdes,
     but are named _ossl_old_des_*.  Finally, add macros that map the
     des_* symbols to the corresponding _ossl_old_des_* if libdes
     compatibility is desired.  If OpenSSL 0.9.6c compatibility is
     desired, the des_* symbols will be mapped to DES_*, with one
     exception.

     Since we provide two compatibility mappings, the user needs to
     define the macro OPENSSL_DES_LIBDES_COMPATIBILITY if libdes
     compatibility is desired.  The default (i.e., when that macro
     isn't defined) is OpenSSL 0.9.6c compatibility.

     There are also macros that enable and disable the support of old
     des functions altogether.  Those are OPENSSL_ENABLE_OLD_DES_SUPPORT
     and OPENSSL_DISABLE_OLD_DES_SUPPORT.  If none or both of those
     are defined, the default will apply: to support the old des routines.

     In either case, one must include openssl/des.h to get the correct
     definitions.  Do not try to just include openssl/des_old.h, that
     won't work.

     NOTE: This is a major break of an old API into a new one.  Software
     authors are encouraged to switch to the DES_ style functions.  Some
     time in the future, des_old.h and the libdes compatibility functions
     will be disable (i.e. OPENSSL_DISABLE_OLD_DES_SUPPORT will be the
     default), and then completely removed.
     [Richard Levitte]

  *) Test for certificates which contain unsupported critical extensions.
     If such a certificate is found during a verify operation it is 
     rejected by default: this behaviour can be overridden by either
     handling the new error X509_V_ERR_UNHANDLED_CRITICAL_EXTENSION or
     by setting the verify flag X509_V_FLAG_IGNORE_CRITICAL. A new function
     X509_supported_extension() has also been added which returns 1 if a
     particular extension is supported.
     [Steve Henson]

  *) Modify the behaviour of EVP cipher functions in similar way to digests
     to retain compatibility with existing code.
     [Steve Henson]

  *) Modify the behaviour of EVP_DigestInit() and EVP_DigestFinal() to retain
     compatibility with existing code. In particular the 'ctx' parameter does
     not have to be to be initialized before the call to EVP_DigestInit() and
     it is tidied up after a call to EVP_DigestFinal(). New function
     EVP_DigestFinal_ex() which does not tidy up the ctx. Similarly function
     EVP_MD_CTX_copy() changed to not require the destination to be
     initialized valid and new function EVP_MD_CTX_copy_ex() added which
     requires the destination to be valid.

     Modify all the OpenSSL digest calls to use EVP_DigestInit_ex(),
     EVP_DigestFinal_ex() and EVP_MD_CTX_copy_ex().
     [Steve Henson]

  *) Change ssl3_get_message (ssl/s3_both.c) and the functions using it
     so that complete 'Handshake' protocol structures are kept in memory
     instead of overwriting 'msg_type' and 'length' with 'body' data.
     [Bodo Moeller]

  *) Add an implementation of SSL_add_dir_cert_subjects_to_stack for Win32.
     [Massimo Santin via Richard Levitte]

  *) Major restructuring to the underlying ENGINE code. This includes
     reduction of linker bloat, separation of pure "ENGINE" manipulation
     (initialisation, etc) from functionality dealing with implementations
     of specific crypto iterfaces. This change also introduces integrated
     support for symmetric ciphers and digest implementations - so ENGINEs
     can now accelerate these by providing EVP_CIPHER and EVP_MD
     implementations of their own. This is detailed in crypto/engine/README
     as it couldn't be adequately described here. However, there are a few
     API changes worth noting - some RSA, DSA, DH, and RAND functions that
     were changed in the original introduction of ENGINE code have now
     reverted back - the hooking from this code to ENGINE is now a good
     deal more passive and at run-time, operations deal directly with
     RSA_METHODs, DSA_METHODs (etc) as they did before, rather than
     dereferencing through an ENGINE pointer any more. Also, the ENGINE
     functions dealing with BN_MOD_EXP[_CRT] handlers have been removed -
     they were not being used by the framework as there is no concept of a
     BIGNUM_METHOD and they could not be generalised to the new
     'ENGINE_TABLE' mechanism that underlies the new code. Similarly,
     ENGINE_cpy() has been removed as it cannot be consistently defined in
     the new code.
     [Geoff Thorpe]

  *) Change ASN1_GENERALIZEDTIME_check() to allow fractional seconds.
     [Steve Henson]

  *) Change mkdef.pl to sort symbols that get the same entry number,
     and make sure the automatically generated functions ERR_load_*
     become part of libeay.num as well.
     [Richard Levitte]

  *) New function SSL_renegotiate_pending().  This returns true once
     renegotiation has been requested (either SSL_renegotiate() call
     or HelloRequest/ClientHello received from the peer) and becomes
     false once a handshake has been completed.
     (For servers, SSL_renegotiate() followed by SSL_do_handshake()
     sends a HelloRequest, but does not ensure that a handshake takes
     place.  SSL_renegotiate_pending() is useful for checking if the
     client has followed the request.)
     [Bodo Moeller]

  *) New SSL option SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION.
     By default, clients may request session resumption even during
     renegotiation (if session ID contexts permit); with this option,
     session resumption is possible only in the first handshake.

     SSL_OP_ALL is now 0x00000FFFL instead of 0x000FFFFFL.  This makes
     more bits available for options that should not be part of
     SSL_OP_ALL (such as SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION).
     [Bodo Moeller]

  *) Add some demos for certificate and certificate request creation.
     [Steve Henson]

  *) Make maximum certificate chain size accepted from the peer application
     settable (SSL*_get/set_max_cert_list()), as proposed by
     "Douglas E. Engert" <deengert@anl.gov>.
     [Lutz Jaenicke]

  *) Add support for shared libraries for Unixware-7
     (Boyd Lynn Gerber <gerberb@zenez.com>).
     [Lutz Jaenicke]

  *) Add a "destroy" handler to ENGINEs that allows structural cleanup to
     be done prior to destruction. Use this to unload error strings from
     ENGINEs that load their own error strings. NB: This adds two new API
     functions to "get" and "set" this destroy handler in an ENGINE.
     [Geoff Thorpe]

  *) Alter all existing ENGINE implementations (except "openssl" and
     "openbsd") to dynamically instantiate their own error strings. This
     makes them more flexible to be built both as statically-linked ENGINEs
     and self-contained shared-libraries loadable via the "dynamic" ENGINE.
     Also, add stub code to each that makes building them as self-contained
     shared-libraries easier (see README.ENGINE).
     [Geoff Thorpe]

  *) Add a "dynamic" ENGINE that provides a mechanism for binding ENGINE
     implementations into applications that are completely implemented in
     self-contained shared-libraries. The "dynamic" ENGINE exposes control
     commands that can be used to configure what shared-library to load and
     to control aspects of the way it is handled. Also, made an update to
     the README.ENGINE file that brings its information up-to-date and
     provides some information and instructions on the "dynamic" ENGINE
     (ie. how to use it, how to build "dynamic"-loadable ENGINEs, etc).
     [Geoff Thorpe]

  *) Make it possible to unload ranges of ERR strings with a new
     "ERR_unload_strings" function.
     [Geoff Thorpe]

  *) Add a copy() function to EVP_MD.
     [Ben Laurie]

  *) Make EVP_MD routines take a context pointer instead of just the
     md_data void pointer.
     [Ben Laurie]

  *) Add flags to EVP_MD and EVP_MD_CTX. EVP_MD_FLAG_ONESHOT indicates
     that the digest can only process a single chunk of data
     (typically because it is provided by a piece of
     hardware). EVP_MD_CTX_FLAG_ONESHOT indicates that the application
     is only going to provide a single chunk of data, and hence the
     framework needn't accumulate the data for oneshot drivers.
     [Ben Laurie]

  *) As with "ERR", make it possible to replace the underlying "ex_data"
     functions. This change also alters the storage and management of global
     ex_data state - it's now all inside ex_data.c and all "class" code (eg.
     RSA, BIO, SSL_CTX, etc) no longer stores its own STACKS and per-class
     index counters. The API functions that use this state have been changed
     to take a "class_index" rather than pointers to the class's local STACK
     and counter, and there is now an API function to dynamically create new
     classes. This centralisation allows us to (a) plug a lot of the
     thread-safety problems that existed, and (b) makes it possible to clean
     up all allocated state using "CRYPTO_cleanup_all_ex_data()". W.r.t. (b)
     such data would previously have always leaked in application code and
     workarounds were in place to make the memory debugging turn a blind eye
     to it. Application code that doesn't use this new function will still
     leak as before, but their memory debugging output will announce it now
     rather than letting it slide.

     Besides the addition of CRYPTO_cleanup_all_ex_data(), another API change
     induced by the "ex_data" overhaul is that X509_STORE_CTX_init() now
     has a return value to indicate success or failure.
     [Geoff Thorpe]

  *) Make it possible to replace the underlying "ERR" functions such that the
     global state (2 LHASH tables and 2 locks) is only used by the "default"
     implementation. This change also adds two functions to "get" and "set"
     the implementation prior to it being automatically set the first time
     any other ERR function takes place. Ie. an application can call "get",
     pass the return value to a module it has just loaded, and that module
     can call its own "set" function using that value. This means the
     module's "ERR" operations will use (and modify) the error state in the
     application and not in its own statically linked copy of OpenSSL code.
     [Geoff Thorpe]

  *) Give DH, DSA, and RSA types their own "**_up_ref()" function to increment
     reference counts. This performs normal REF_PRINT/REF_CHECK macros on
     the operation, and provides a more encapsulated way for external code
     (crypto/evp/ and ssl/) to do this. Also changed the evp and ssl code
     to use these functions rather than manually incrementing the counts.

     Also rename "DSO_up()" function to more descriptive "DSO_up_ref()".
     [Geoff Thorpe]

  *) Add EVP test program.
     [Ben Laurie]

  *) Add symmetric cipher support to ENGINE. Expect the API to change!
     [Ben Laurie]

  *) New CRL functions: X509_CRL_set_version(), X509_CRL_set_issuer_name()
     X509_CRL_set_lastUpdate(), X509_CRL_set_nextUpdate(), X509_CRL_sort(),
     X509_REVOKED_set_serialNumber(), and X509_REVOKED_set_revocationDate().
     These allow a CRL to be built without having to access X509_CRL fields
     directly. Modify 'ca' application to use new functions.
     [Steve Henson]

  *) Move SSL_OP_TLS_ROLLBACK_BUG out of the SSL_OP_ALL list of recommended
     bug workarounds. Rollback attack detection is a security feature.