source: trunk/Packages/synapse/README

Changes in 3.2.2 release
========================

The bignum code has been updated to the latest version, which is no less
ghastly than before but somewhat faster, particularly on x86 systems.

Changes in 3.2.1 release
========================

The zlib code has exhibited a number of security vulnerabilities, both in the
rather dated 1998-vintage zlib 1.1.x code and the much newer zlib 1.2.x code.
In order to avoid exposing cryptlib users to any new technology until it's
proven itself in the field, cryptlib has stayed with the old zlib code base,
which has avoided problems arising from the vulnerabilities found in the 1.2.x
code.  However, the 1.2.x code base does have a number of advantages over the
old 1.1.4 code, including better compression, much faster compression and
decompression, and significantly less memory consumption, particularly for
decompression.  Another potential problem with 1.1.4 is that there may be as
yet undiscovered problems in it that no-one's found yet because they're
concentrating on the newer 1.2.x instead.

Because zlib 1.2.x includes a significant rewrite of a fair portion of the
code, it's not possible to simply switch from one version of zlib to the other
at the flip of a compiler switch.  Based on user feedback, the preferred
behaviour is to have cryptlib move from the old 1.1.x to the current 1.2.x
code base, so cryptlib 3.2.1 uses zlib 1.2.3 (those values are pure
coincidence).  Users concerned about security issues should disable the
compression code (undefine USE_COMPRESSION in cryptini.h) unless it's actually
required.

Support of autoproxy configuration (under operating systems that implement
automatic network proxy discovery) has been added.  This includes support for
SOCKS-style HTTP proxies as well as the originally implemented HTTP-
application-gateway HTTP proxies.

Several minor portability issues in 3.2 have been fixed, among other things a
workaround for a problem in gcc 4.0 that prevented the code from compiling.

Changes in 3.2 release
======================

The version that would have been released as 3.11 has been released as 3.2 to
avoid version-numbering confusion. Future releases will use a three-digit
numbering scheme x.y.z, where each of x, y, and z will constitute a single
digit, to avoid confusion over whether the '11' in '3.11' is 11 or 0.11.

The network timeout handling has been split into distinct read and write
timeouts to allow more flexible configuration of network data handling,
typically very small read timeouts to ensure that cryptlib never blocks while
waiting for data, and slightly larger write timeouts to ensure that all data
is completely written when requested.

Client authentication for sessions that support this (only the secure data
sessions, SSH and SSL/TLS, use client authentication) has been changed
slightly.  In previous releases, cryptlib would always allow the client's
authentication and complete the handshake, leaving it to the caller to shut
down the session later on if there was a problem.  As of this release, server
sessions will return a CRYPT_ENVELOPE_RESOURCE in the same way that envelopes
do when you activate them, to indicate that you need to provide further
information in order to continue.  If you receive this status when you try to
activate a server session, you can read the user details via the
CRYPT_SESSINFO_USERNAME and CRYPT_SESSINFO_PASSWORD attributes, and either
comfirm the authentication by setting the CRYPT_SESSINFO_AUTHRESPONSE to true
or deny it by setting it to false.  You then re-activate the session, and
cryptlib will complete the handshake.  This allows you to check the
authentication details before allowing the handshake to continue.  If you'd
prefer the previous behaviour, just set CRYPT_SESSINFO_AUTHRESPONSE to true
before you activate the session and cryptlib will handle confirmation of
authentication itself.

Because the handshake is now controlled by the server rather than the client
(that is, cryptlib won't sit in a loop reading all data from the client, but
will return to the caller to await further input), it's possible that the
client will send further session-control information after the handshake has
completed.  To respond to any additional information that the client may send,
your server should try and pop client data before it sends any of its own
data.  If the server sends its data first, the client may interpret the
server's data as an (invalid) response to a request that it has in progress.
The manual contains more details on the new client authentication and server
data handling process.

Starting with releases after 3.2, the attribute cursor manipulation functions
will be unified so that they work identically for all container objects that
maintain lists of attributes (most of this interface is already available in
3.2 for forwards-compatibility reasons).  This means that the three attribute-
list mechanisms CRYPT_CERTINFO_CURRENT_EXTENSION,
CRYPT_CERTINFO_CURRENT_FIELD, and CRYPT_CERTINFO_CURRENT_COMPONENT (for
certificates), CRYPT_ATTRIBUTE_CURRENT_GROUP (formerly
CRYPT_ENVINFO_CURRENT_COMPONENT) for envelopes, and the future
CRYPT_ATTRIBUTE_CURRENT_GROUP for session objects will be folded into the
single mechanism CRYPT_ATTRIBUTE_CURRENT_GROUP, CRYPT_ATTRIBUTE_CURRENT, and
CRYPT_ATTRIBUTE_CURRENT_INSTANCE.  This provides both a unified, consistent
interface across the different object types and allows envelope and session
attributes to be handled in the same way as certificates currently are.  For
example under the older cryptlib 3.1 behaviour envelope signatures were
implicitly selected by reading the signature attribute, which changed the
current selection of the signature result and extra signature information
attributes.  From releases after this one envelope signature attributes are
grouped by signature, so selecting the first signature explicitly selects the
associated signature key, signature result, and extra signature information
attributes, selecting the second signature explicitly selects the associated
information for that, and so on.  In addition reading this information under
3.1 and earlier was something of a hunt-and-peck approach that required
reading each possible attribute in turn to see whether it was present.  The
attribute cursor interface allows each present attribute to be enumerated
using the attribute cursor, with cryptlib doing the work of sorting out what's
present and what isn't.  Finally, cryptlib 3.2 allows more session attributes
to be accessed by the caller, which requires the use of attribute cursors
since many of them are composite values or multi-valued attributes.

As part of this unification of mechanisms, the CRYPT_ENVINFO_CURRENT_COMPONENT
attribute has been renamed CRYPT_ATTRIBUTE_CURRENT_GROUP for compatibility
with future cryptlib releases.  For certificates, cryptlib 3.11 supports both
the legacy CRYPT_CERTINFO_CURRENT_xxx attributes and the new
CRYPT_ATTRIBUTE_CURRENT_xxx ones, so you can either keep using the
certificate-specific cursor attributes for now or move to the universal cursor
attributes for compatibility with future versions.  All that's necessary to
upgrade is a search and replace in your code to rename the attributes.  More
information on the unified attribute cursor mechanism is given in the
Introduction section of the manual.

The various situation-specific SSH attributes like
CRYPT_SESSINFO_SSH_SUBSYSTEM and CRYPT_SESSINFO_SSH_PORTFORWARD have been
replaced with more generic forms for compatibility with future cryptlib
versions, which allow more complete control over SSH channel types and
parameters.  See the manual for details on creating and controlling SSH
channels using the new attributes.

The SHA-2 algorithm is now enabled by default (previously it had to be
explicitly enabled via a compile-time option).  Note that this algorithm isn't
supported by most software and standards, so using it in anything other than
closed environments will lead to interoperability problems.  If you want to
use SHA-2, you should test it against any other software that you need to
interoperate with to ensure that it can handle this algorithm.

The Unix randomness slow poll now has an early-out mechanism that avoids the
full heavyweight slow poll if sufficient entropy is available without it.
This typically occurs on systems with both built-in randomness sources and
either a kernel statistics interface or an advanced /proc interface, which
provide the same data as the full slow poll at a fraction of the overhead.

Changes in 3.11 beta 1
======================

As noted in the 3.1 release notes, releases after 3.1 update the cryptlib 2.x
legacy functions cryptImport/ExportKey(Ex), cryptCreate/CheckSignature(Ex),
and cryptQueryObject to take a length parameter as their second argument.
These functions never took a length parameter in their original form because
they dealt with encoded signature/key exchange data, for which it's always
possible to determine the length by parsing it.  However, this makes checking
of user-supplied parameters difficult - cryptlib always knows how much output
it'll produce, but it can't be sure that the user wants (or expects) that much
output, particularly if they've forgotten to perform a length check before
exporting a key or creating a signature.  In order to avoid this problem,
releases after 3.1 make this minor API change, which fixes the last of the old
cryptlib 2.x functions.  Since most people use the higher-level API functions,
this change shouldn't affect the majority of users.

In addition to the cryptlib 2.x legacy functions, two standard data-returning
functions cryptExportCert and the rarely-used cryptGetCertExtension now also
take a length parameter to specify the maximum buffer size.  This brings them
into line with the (updated) legacy functions, and allows cryptlib to perform
better checking of user-supplied parameters, particularly in the case of
languages like Visual Basic and Java where the mapping of language-native data
types to the blocks of memory used by a C-language library can be tricky.

Changes in 3.1 release
======================

The final release contains mostly minor tweaks based on user feedback from the
3.1 final betas, with no noticeable external changes.  Internally, the HTTP
engine has been significantly improved, TLS 1.1 is now supported (although at
release time there were no other known implementations of this to test
against), the BeOS port has been re-done to handle the current state of the OS
using GNU development tools instead of the original Be ones (thanks to Simon
Taylor for providing access to his system to do the work on), and the
perpetual tweaking of the networking subsystem to handle OS-specific quirks
has continued.

Changes in 3.1 beta 5
=====================

This release contains an extensive rewrite of the network-layer code to handle
various quirks and system-specific peculiarities in sockets implementations.
For this reason it's strongly recommended that if you're using secure sessions
or networking, you move to this version as quickly as possible.  In addition
the code is now IPv6-enabled as part of the rewrite.  In connection with this,
it's now possible to specify connect and network timeouts at the per-session
level rather than only as a cryptlib-wide option. To do this, set the timeout
value giving a specific session as the target object rather than CRYPT_UNUSED
for all of cryptlib:

  cryptSetAttribute( cryptSession, CRYPT_OPTION_NET_TIMEOUT, timeout );

All of the session objects now provide extensive text diagnostics of errors
via the CRYPT_ATTRIBUTE_INT_ERRORMESSAGE attribute.  This means that if a
session encounters a problem in communicating with another system, the
extended error message will usually provide additional information that goes
beyond the standard error code.

Support for SSHv2 subsystems such as SFTP has been added via the
CRYPT_SESSINFO_SSH_SUBSYSTEM attribute, and support for SSL/TLS with shared
keys (e.g. passwords) has been added.  This allows secure SSL with mutual
authentication of both client and server, without the need for either
expensive server certificates or the complexity of client certificates.  In
addition the shared-key mechanism is significantly faster than operations
involving certificates and private keys, and places far less load on both
client and server.

The CRYPT_OPTION_CERT_OBLIVIOUS option introduced in beta 4 has been changed
from a simple boolean value to a multivalued option
CRYPT_OPTION_CERT_COMPLIANCELEVEL, which indicates the level of checking
applied to certificates.  This ranges from CRYPT_COMPLIANCELEVEL_OBLIVIOUS
(almost no checking except for the signature) through to
CRYPT_COMPLIANCELEVEL_PKIX_FULL (full compliance with PKIX certificate
standards).  The default is CRYPT_COMPLIANCELEVEL_STANDARD, which provides a
reduced level of checking for compliance with other certificate software.

The CRYPT_ERROR_BUSY status has been renamed to CRYPT_ERROR_TIMEOUT to make it
more consistent with conditions like network timeouts, which is where it most
commonly occurs.

Some of the database keyset types have been renamed to make the names more
consistent.  The three types are now CRYPT_KEYSET_ODBC for databases accessed
through an ODBC interface, CRYPT_KEYSET_DATABASE for databases with the
backend-specific interface code compiled into cryptlib, and
CRYPT_KEYSET_PLUGIN for databases accessed via the cryptlib database plugin
interface.  The CRYPT_KEYSET_DATABASE type can be selected in the usual manner
at compile time with the USE_xxx compile options, see the manual for more
details on databases and compile options.

Changes in 3.1 beta 4
=====================

The changes in this release are mostly to fix a variety of minor problems that
cropped up in beta 3 related to portability across different systems.  One
notable change is the inclusion of the CRYPT_OPTION_CERT_OBLIVIOUS
configuration option.  This option is present in order to allow cryptlib to
handle the large number of broken certificates in use.  Enabling this option
(setting it to TRUE) will cause cryptlib to ignore all certificate extensions
except key usage and the CA flag, making it compatible with other PKI
software.  A future version of cryptlib will replace this option with a
multivalue selection allowing a choice of checking ranging from very little
(in order to be compatible with existing software and to process existing
certificates) through to full PKIX compliance (which will, however, reject a
large number of certificates in use today).

Changes in 3.1 beta 3
=====================

The way in which GeneralNames and DNs in GeneralNames are selected has
changed.  Until now GeneralNames were selected with:

  cryptSetAttribute( cryptCert, attributeID, CRYPT_UNUSED );

which selected the GeneralName with the given attributeID and also updated the
attribute cursor.  Selecting the DN within the GeneralName required a second
selection to specify the DN.  This special-case behaviour was inconsistent
with the way other attributes were handled, and caused some confusion for
users, as well as obscuring the fact that the attribute cursor was being
updated as it would be for an explicit attribute selection.  Beta 3 changes
the selection mechanism so that it matches that used for all other attributes:

  cryptSetAttribute( cryptCert, CRYPT_CERTINFO_CURRENT_EXTENSION,
                     attributeID );

(or whatever alternative attribute cursor mechanism is preferred).  In
addition, it's no longer necessary to perform a two-stage select for DNs in
GeneralNames, since these are now implicitly selected by selecting the
GeneralName that contains them.  This means that handling of GeneralName
attributes is now consistent with all other attribute types.

Because GeneralNames are now automatically selected, you need to be careful
when working with a DN such as the certificate subject name after selecting a
GeneralName, since a GeneralName contains its own DN which will be the one
that's accessed after selecting the GeneralName.  For example if you set an
email address (which is part of the GeneralName), this will select the
GeneralName, and any subsequent DN accesses will refer to the DN inside the
GeneralName rather than the subject name DN.  It's good practice to move back
to the subject name with:

  cryptSetAttribute( cryptCert, CRYPT_CERTINFO_SUBJECTNAME,
                     CRYPT_UNUSED );

after working with a GeneralName to make sure that you're referring to the
correct DN.

Microsoft broke the Jet (MS Access) database engine at around SP4 (opinions on
the exact time vary).  This version, or newer versions, may be installed
automatically when applying service packs or installing other software
(different broken versions of Jet are installed in Win2K SP2 and SP3, for
example, and it's built into Windows XP).  The symptoms are that when running
the self-test you'll get an error message "Cannot open any more tables" even
though no tables are open, or more seriously a deadlock inside the VC++
runtime file access routines.  The details of this problem are discussed in
Microsoft Knowledge Base Article 304536, Microsoft Knowledge Base Article
282010, and Microsoft Knowledge Base Article 239114.

The Microsoft-recommended fix is to upgrade to Jet SP6.  This typically
requires downloading and installing the SP6 upgrade from the Microsoft
Download Centre, simply installing recent MDAC components won't work since
there were no Jet components included after MDAC 2.5 SP2, and installing
Office XP or Access 2002 may not work either since the setup programs only
update system files in certain situations and to a certain level.  On the
other hand Jet SP6 may not install unless it detects an appropriate service
pack level on the system.

Non-Microsoft advice is to downgrade to a version of Jet at SP3 or earlier by
replacing msjetoledb40.dll and msjet40.dll with files from JET OLE DB SP3,
since even the latest version (SP6) still appears to exhibit these problems.
A better solution is to avoid the use of Jet entirely, and in particular you
should *never* use Jet in a production environment since it's far too buggy
and unstable to trust live data to (it was probably named Jet because it both
sucks and blows).

Changes in 3.1 beta 2
=====================

The last two type names that didn't follow the pattern xxx_TYPE, CRYPT_ALGO
and CRYPT_MODE, have now been brought into line with the other names, so you
need to change the type names to CRYPT_ALGO_TYPE and CRYPT_MODE_TYPE if you're
using them in your code.

SSHv2 server support is now present.

PKCS #11 devices can now be auto-detected by specifying the device name as
"[Autodetect]" if the device name isn't known.

Pre-connected network sockets can be supplied to sessions as the
CRYPT_SESSINFO_NETWORKSOCKET attribute, rather than having cryptlib
connect/disconnect the socket itself.

The PKCS #15 format has had various features added to it over time (see the
note for 3.0 beta 1), cryptlib has supported these in a read-only manner,
meaning that it creates keysets that are as backwards-compatible as possible
with old versions of the code that used earlier versions of PKCS #15.  In
order to make some features like certificate update in a keyset possible, it's
necessary to write keyset fields from newer versions of PKCS #15.  All betas
after this one will write these newer fields in order to enable advanced
features of PKCS #15.  Anyone still using very old betas should upgrade to a
current version in order to take advantage of the advanced features supported
by newer-format PKCS #15 keysets.

Changes in 3.1 beta 1
=====================

Beta 1 includes support for PGP and OpenPGP, which is handled in standard
envelope fashion with a data format of CRYPT_FORMAT_PGP.  Note that there are
a number of incompatible and semi-compatible variants of PGP in existence,
with the major types being PGP 2.x, PGP 5.x, NAI PGP, GPG, and the ckt PGP
builds, and many interesting lesser variations such as the disastry builds
that take PGP 2.x and update it to support OpenPGP ciphers without actually
being OpenPGP.  An overview of different PGP versions can be found at
http://rmarq.pair.com/pgp/, and detailed information on incompatibilities can
be found at http://home.arcor.de/kraven/pgp/pgpindex.html.  cryptlib has been
tested against most normal versions of PGP, but there will certainly be
versions out there that produce data that cryptlib can't read (one example
being a version that uses random key IDs in encrypted data in order to obscure
who the data is encrypted for, which also makes it impossible to decrypt with
any normal PGP version).  If you run into something produced by one of these
oddball versions, please send me a copy so that I can determine whether it's
worth adding custom code to support it.

The PGP data format imposes a number of restrictions on the standard
enveloping process, which include:

 - Use of nested signed data is discouraged unless you add an intermediate
   layer of encryption, since PGP doesn't directly handle nested signatures.
 - When signing data, the nested content type can only be the default (data)
   for the reason given above.
 - It's not possible to mix password and public-key key exchange actions in
   the same message.  Similarly, it's not possible to have more than one
   password key exchange action in one message.
 - Handling of separate hashing for detached signatures is somewhat peculiar
   and requires special care.

Changes in 3.0 release
======================

cryptlib 3.1 will introduce a new function cryptFlushData() to replace the
current practice of calling cryptPushData() will all-null parameters, a change
required for languages like Delphi and VB that don't handle C null pointers
too well.  This function is already present in 3.0 for forwards-compatibility
purposes, it's recommended that you update your code to call cryptFlushData()
in place of cryptPushData() with null parameters (although the existing
cryptPushData() mechanism will still work).

The Unix randomness-gathering code will now check for and use EGD/PRNGD if
they're available.

The requirement that cryptlib be built via a network share under Windows has
been removed.

HTTP keyset access (CRYPT_KEYSET_HTTP) formerly required that the keyset be
opened without a name being given, with the full URL being specified as the
key ID to retrieve keys.  This was both somewhat inconsistent with the other
keyset types, and didn't work well with persistent connections, for example
where multiple certificates were being read from a single server.  This has
been changed so that the server URL is given when the keyset is opened as it
is for other keyset types, and a key ID is given when reading individual keys.
When reading keys with a fixed URL (with no per-key ID), the special ID
"[none]" can be used to indicate that the server URL points directly at the
certificate.  In the simplest case the previous usage:

  cryptKeysetOpen( &cryptKeyset, CRYPT_UNUSED, CRYPT_KEYSET_HTTP, NULL,
                   CRYPT_KEYOPT_READONLY );
  cryptGetPublicKey( cryptKeyset, &cryptCert, CRYPT_KEYID_NAME,
                     "http://www.server.com/cert.der" );

now becomes:

  cryptKeysetOpen( &cryptKeyset, CRYPT_UNUSED, CRYPT_KEYSET_HTTP,
                   "http://www.server.com/cert.der", CRYPT_KEYOPT_READONLY );
  cryptGetPublicKey( cryptKeyset, &cryptCert, CRYPT_KEYID_NAME,
                     "[none]" );

Reading multiple certificates, for example via a CGI interface on the server,
is done with:

  cryptKeysetOpen( &cryptKeyset, CRYPT_UNUSED, CRYPT_KEYSET_HTTP,
                   "http://www.server.com/certstore.cgi",
                   CRYPT_KEYOPT_READONLY );
  cryptGetPublicKey( cryptKeyset, &cryptCert, CRYPT_KEYID_NAME,
                     "user1" );
  cryptGetPublicKey( cryptKeyset, &cryptCert, CRYPT_KEYID_NAME,
                     "user2" );
  cryptGetPublicKey( cryptKeyset, &cryptCert, CRYPT_KEYID_NAME,
                     "user3" );

Changes in 3.0 final beta
=========================

The cryptlib 3.0 final release divides the network timeout parameter into two
parts, a CRYPT_OPTION_NET_CONNECTTIMEOUT which is applied during the
connection setup process and a CRYPT_OPTION_NET_TIMEOUT which is applied
during reads and writes (although in practice writes are almost always
instantaneous).  This means that it's now possible to avoid nonblocking I/O if
required.

Use of SSL/TLS client certificates is now enabled.

The final version of the S/MIME PasswordRecipientInfo (PWRI) RFC contained a
change in the way the key wrap algorithm is identified.  The cryptlib final
release produces a PWRI that follows the final RFC, but will also read the
older format produced by earlier versions of cryptlib.  If it's necessary to
generate PWRI data in the old format, you can change the "#if 1" in
keymgmt/asn1objs.c to "#if 0" to produce the older format.

Support for extended CMP user configurability via PKIUser objects has been
added, this allows user details to be pre-configured at the CA rather than
requiring the user to know them.

Changes in 3.0 beta 6
=====================

Beta 6 reduces the plethora of key generation functions by allowing the
keysize to be specified in the more standard way of setting the corresponding
attribute rather than having extraneous ...Ex() versions of the function that
parallel the standard form.  If you were previously using the standard or
asynchronous ...Ex() functions:

        cryptGenerateKeyEx( cryptContext, keySize );

you can change to the newer form with a simple cut and paste:

        cryptSetAttribute( cryptContext, CRYPT_CTXINFO_KEYSIZE, keySize );
        cryptGenerateKey( cryptContext );

The table format for certificate stores has changed slightly to accomodate the
requirements of CMP servers (specifically the fact that they can cause things
to fail at inopportune moments), in order to handle this you need to recreate
the cert store (drop the previous table and create a new one with
CRYPT_KEYOPT_CREATE), if you need to move information across then create a new
table and insert into <newTable> select * from <oldTable> to copy the existing
certificates across.  The newly-created table will support the storing of
extra information needed to handle restarts during CMP operations.

Support for the proposed AES algorithm is now included.  Since this hasn't
been finalised by NIST yet, it is strongly recommended that you not use it
until the AES FIPS has been published.  cryptlib will implement the form
presented in the final FIPS, which may be incompatible with the current
version.

SSL and TLS support (both client and server) is now enabled, thanks to
endergone Zwiebeltuete for his help in getting this going.

Changes in 3.0 beta 5
=====================

Beta 5 removes the configuration options CRYPT_OPTION_FIXSTRINGS and
CRYPT_OPTION_CHECKENCODING since they shouldn't be needed any more as the
ASN.1 code has been modified to handle all but the most broken certificates
out there.

Under Unix the library is now called libcl rather than libcrypt to reduce
problems with naming conflicts on some systems.

Support for the creation of one-step certificates that automatically work for
any purpose has been added and is enabled by setting the certificate's
CRYPT_CERTINFO_XYZZY attribute, more details are given in the manual.

OCSP, TSP, and SSH server support is now enabled (this complements the OCSP,
TSP, and SSH client support added in beta 3 and beta 4).

Changes in 3.0 beta 4
=====================

Beta 4 removes the need to use the TCP4U library for the network interface and
should now have networking enabled automatically on Unix and Windows systems.
With the enhanced networking support comes support for SSH secure sessions and
CMP (Certificate Mismanagement Protocol).

The creation of arbitrary-format DN's containing any sort of component in any
order or combination is now supported via the CRYPT_CERTINFO_DN attribute.

Changes in 3.0 beta 3
=====================

Beta 3 required one unfortunate change to the object creation functions which
involves the addition of a parameter that specifies the user who is to own the
object.  For cryptlib 3.0 this means that as the second parameter for the
object creation functions (cryptCreateContext, cryptKeysetOpen,
cryptCreateEnvelope, etc etc) you should specify CRYPT_UNUSED for forwards
compatibility with cryptlib 3.1 and later releases.  The user parameter has
existed since the first 3.0 releases but until now has been hidden beneath the
cryptlib API, unfortunately it's necessary to un-hide this parameter, which is
required by certain 3.1 features such as the CA management functionality.

The use of multiple certificates with the same DN/email address but different
key usage types is now supported (previous versions treated the appearance of
multiple certs issued to the same person as a duplicate cert problem, either
interpretation is valid but popular opinion seems to be that the beta 3
behaviour is better).  In order to support this in public key keysets you need
to recreate them (drop the previous table and create a new one with
CRYPT_KEYOPT_CREATE), if you need to move certificates across then create a
new table and insert into <newTable> select * from <oldTable> to copy the
existing certificates across.  The newly-created table will support the
storing of multiple identical certificates.

In order to take advantage of this capability with encrypted enveloping, you
need to add the recipient's public key with the CRYPT_ENVINFO_RECIPIENT option
that will allow cryptlib to automatically select the most appropriate
certificate.  The selection of an appropriate signature checking certificate
is handled automatically.

Handling of detached signatures with externally-supplied hash values is now
supported, see the manual for details.  This allows signature checking for
arbitrary data without it having to be hashed via the envelope.

Certificate revocation checking via OCSP is now supported if you can manage to
find an OCSP responder anywhere.  Timestamping of signatures is also
supported, although you'll have even less chance of locating a TSA than an
OCSP responder.

Support for mSQL has been dropped since MySQL is now GPL'd and provides far
more functionality than mSQL (even cryptlib 2.x was pushing the limits of
mSQL, with cryptlib 3.x it's not really sufficient any more).

Changes in 3.0 beta 2
=====================

Beta 2 has three minor API changes over beta 1 that are intended to clarify
areas that have caused problems for users in the past.

The first change is that the third, usually unnecessary, parameter for
cryptCreateContext() has been eliminated:

  cryptCreateContext( &cryptContext, cryptAlgo );

Only the conventional-encryption algorithms required the encryption mode
parameter, the default is CBC but if another value is required you can specify
it using the CRYPT_CTXINFO_MODE attribute:

  cryptSetAttribute( cryptContext, CRYPT_CTXINFO_MODE, cryptMode );

This attribute isn't required for anything except conventional encryption
algorithms, and even then it's only required for the use of encryption modes
other than the default mode of CBC.  This change simplifies the creation of
contexts, since there's no longer any need to juggle CRYPT_USE_DEFAULT,
CRYPT_MODE_PKC, CRYPT_MODE_NONE, or any of the other special cases that were
used for algorithms that don't have different encryption modes.

The second change (which probably won't affect most users since it's rather
obscure) is that previously when loading raw public keys with the
CRYPT_PKCINFO_RSA or CRYPT_PKCINFO_DLP structure the attribute used was
CRYPT_CTXINFO_KEY and the length was given as CRYPT_UNUSED.  This required a
number of special-case checks in the code and made error-checking of user-
supplied data impossible because it wasn't possible to determine how much data
was being passed in.

In beta 2 this attribute now follows the pattern for every other attribute in
that it's necessary to give the structure size (i.e. sizeof(
CRYPT_PKCINFO_xxx)) as the length parameter rather than CRYPT_UNUSED.  In
addition the attribute for key components is now CRYPT_CTXINFO_KEY_COMPONENTS
to make explicit the fact that what's being loaded is a composite structure
containing multiple components and not just a single byte string as with
CRYPT_CTXINFO_KEY.

Similarly, the length parameter when using cryptEncrypt()/cryptDecrypt() for
public-key encryption is now the key/data length in bytes rather than
CRYPT_UNUSED, following the standard pattern of requiring an actual length
rather than using magic values.

The third change is that the functionality of cryptCreateEnvelopeEx() has been
folded into cryptCreateEnvelope(), which now takes a second parameter
specifying the type of envelope to create

  cryptCreateEnvelope( &cryptEnvelope, formatType );

The envelope buffer size can optionally be specified with the
CRYPT_ATTRIBUTE_BUFFERSIZE attribute once the envelope has been created:

  cryptSetAttribute( cryptEnvelope, CRYPT_ATTRIBUTE_BUFFERSIZE, size );

although this should only be necessary in special cases when enveloping
larger-than-usual data quantities.  This change both simplifies the interface
and makes it easier for cryptlib to efficiently handle resources, since it can
allocate a small envelope buffer when enveloping begins rather than having to
create a one-size-fits-all one on envelope creation.

In addition to these changes, beta 2 includes the ability to read the label
for the private key which is required for de-enveloping data, so you can use
the key name in prompts when asking the user for a password.  You can do this
with:

  char label[ CRYPT_MAX_TEXTSIZE + 1 ];
  int labelLength;

  cryptGetAttributeString( cryptEnvelope, CRYPT_ENVINFO_PRIVATEKEY_LABEL,
                           label, &labelLength );
  label[ labelLength ] = '\0';

See the manual for more information on this.

Changes in 3.0 beta 1
=====================

cryptlib 3.0 features a large number of improvements over 2.1, the most
visible one being that there is now a unified object model that applies to all
cryptlib objects, so that the old object-type-specific functions like
cryptGetCertComponentString() and cryptSetEnvelopeComponentNumeric() have been
replaced by cryptSetAttribute(), which works across all object types.  This
means that cryptlib 3.0 has a much simpler interface than 2.1 did (even with
all the features added in 3.0, the manual is 25 pages shorter than the 2.1
manual).

Backwards compatibility with 2.1 is maintained through the use of the 2.1
include file capi.h, which contains macros that map the 3.0 functions and
attributes back to 2.1 versions.  Three of the more obscure functions don't
translate cleanly, these are documented at the start of capi.h.  If you've got
existing 2.1 code then it should work with 3.0 with a recompile.

This is a beta release of the code, some sections are still subject to change
because the standards they are based on haven't been finalised yet. The most
obvious one is the PKCS #15 keyset format, when the cryptlib code was frozen
the PKCS #15 file format existed only in an informal manner so that some of
the data formatting used by cryptlib is speculative and will probably change
when the standard stabilises.  What this means in practice is that you
shouldn't store any long-term keys in file keysets since the format will have
to change in the future to track changes in the standard.

To a lesser extent, the compressed-data and password-protected-data formats
are based on S/MIME drafts that may also be changed at some point.

Finally, the AS/400, MVS, and VM/CMS versions have somewhat specialised
requirements (some of this is covered in the manual), please contact me before
trying to do anything with these versions.