Andy Polyakov [Fri, 28 May 2004 22:18:48 +0000 (22:18 +0000)]
Unified hpux-shared rule. Verified with both 32- and 64-bit builds and
both vendor and GNU compilers. ./engine shared build are still busted.
I mean always were...
Andy Polyakov [Tue, 25 May 2004 20:31:03 +0000 (20:31 +0000)]
Framework for glueing BIO layer and Win32 compiler run-time. Goal is to
make it possible to produce for a unified binary build, which can be
used with a variety of Win32 compilers.
Andy Polyakov [Thu, 20 May 2004 21:18:09 +0000 (21:18 +0000)]
Final API adaptation. Final, "all openssl" performance numbers [not mixture
of different implementations]. Real-life performance improvement is rated
at 2-3x, not 6x as preliminary announced.
Geoff Thorpe [Mon, 17 May 2004 19:26:06 +0000 (19:26 +0000)]
After the latest round of header-hacking, regenerate the dependencies in
the Makefiles. NB: this commit is probably going to generate a huge posting
and it is highly uninteresting to read.
Geoff Thorpe [Mon, 17 May 2004 19:14:22 +0000 (19:14 +0000)]
Deprecate the recursive includes of bn.h from various API headers (asn1.h,
dh.h, dsa.h, ec.h, ecdh.h, ecdsa.h, rsa.h), as the opaque bignum types are
already declared in ossl_typ.h. Add explicit includes for bn.h in those C
files that need access to structure internals or API functions+macros.
Geoff Thorpe [Mon, 17 May 2004 18:58:47 +0000 (18:58 +0000)]
I can't verify this directly, but recent changes will probably require that
the cryptodev implementation include bn.h directly (when building with
OPENSSL_NO_DEPRECATED that is).
Geoff Thorpe [Mon, 17 May 2004 18:53:47 +0000 (18:53 +0000)]
Deprecate quite a few recursive includes from the ssl.h API header and
remove some unnecessary includes from the internal header ssl_locl.h. This
then requires adding includes for bn.h in four C files.
Geoff Thorpe [Sat, 15 May 2004 18:32:08 +0000 (18:32 +0000)]
Make some more API types opaquely available from ossl_typ.h, meaning the
corresponding headers are only required for API functions or structure
details. This now includes the bignum types and BUF_MEM. Subsequent commits
will remove various dependencies on bn.h and buffer.h and update the
makefile dependencies.
Andy Polyakov [Thu, 6 May 2004 10:36:49 +0000 (10:36 +0000)]
SSE2 accelerated bn_mul_add_words. Code is currently disabled till proper
config and run-time support is added.
PR: 788
Submitted by: <dean@arctic.org>
Reviewed by: <appro>
Bodo Möller [Tue, 4 May 2004 01:15:48 +0000 (01:15 +0000)]
- update from current 0.9.6-stable CHANGES file
- update from current 0.9.7-stable CHANGES file:
Now here we have "CHANGES between 0.9.7e and 0.9.8", and I hope
that all patches mentioned for 0.9.7d and 0.9.7e actually are
in the CVS HEAD, i.e. what is to become 0.9.8.
I have rewritten the 'openssl ca -create_serial' entry (0.9.8)
so that it explains the earlier change that is now listed (0.9.7e).
The ENGINE_set_default typo bug entry has been moved from 0.9.8
to 0.9.7b, which is where it belongs.
Andy Polyakov [Tue, 27 Apr 2004 22:05:50 +0000 (22:05 +0000)]
Improved PowerPC support. Proper ./config support for ppc targets,
especially for AIX. But most important BIGNUM assembler implementation
submitted by IBM.
Submitted by: Peter Waltenberg <pwalten@au1.ibm.com>
Reviewed by: appro
The problem of rsa key-generation getting stuck in a loop for (pointlessly)
small key sizes seems to result from the code continually regenerating the
same prime value once the range is small enough. From my tests, this change
fixes the problem by setting an escape velocity of 3 repeats for the second
of the two primes.
Allow RSA key-generation to specify an arbitrary public exponent. Jelte
proposed the change and submitted the patch, I jiggled it slightly and
adjusted the other parts of openssl that were affected.
New option to 'x509' -next_serial. This outputs the certificate
serial number plus 1 to the output file. Its purpose is to allow
serial number files to be initialized when random serial numbers
are used.
Richard Levitte [Tue, 20 Apr 2004 10:57:07 +0000 (10:57 +0000)]
Print the debug thingies on stderr instead of stdout. If for nothing
else then at least so bc doesn't have problems parsing the output from
bntest :-).
When generating dependencies in the makefiles, generate the reduced
dependencies of the OPENSSL_NO_DEPRECATED mode. This prevents dependencies
being reproduced for "deprecated" header behaviour when a developer doesn't
define the symbol (with the subsequent CVS wars that can ensue).
(oops) Apologies all, that last header-cleanup commit was from the wrong
tree. This further reduces header interdependencies, and makes some
associated cleanups.
Reduce header interdependencies, initially in engine.h (the rest of the
changes are the fallout). As this could break source code that doesn't
directly include headers for interfaces it uses, changes to recursive
includes are covered by the OPENSSL_NO_DEPRECATED symbol. It's better to
define this when building and using openssl, and then adapt code where
necessary - this is how to stay current. However the mechanism exists for
the lethargic.
Andy Polyakov [Sun, 28 Mar 2004 21:27:47 +0000 (21:27 +0000)]
This is essentially Intel 32-bit compiler tune-up. To start with all
available compiler versions generated bogus machine code trying to
compile new crypto/des/cfb_enc.c. Secondly, 8th version defines
__GNUC__ macro, but fails to compile *some* inline assembler correctly.
Note that all versions of icc implement MSC-like _lrot[rl] intrinsic,
which is used now instead of offensive asm. Finally, unnecessary linker
dependencies are eliminated. Most notably dependency from libirc.a
caused trouble at application start-up, if libcrypto.so is linked with
-Bsymbolic (which it is).
Richard Levitte [Thu, 25 Mar 2004 20:09:00 +0000 (20:09 +0000)]
Move the definition of Win32_rename(), since the macro rename gets undefined
in the middle of the code on Windows, and that disrupts operations in functions
later that use rename()...
PR: 853
Geoff Thorpe [Thu, 25 Mar 2004 04:32:24 +0000 (04:32 +0000)]
Adjust various bignum functions to use BN_CTX for variables instead of
locally initialising their own.
NB: I've removed the "BN_clear_free()" loops for the exit-paths in some of
these functions, and that may be a major part of the performance
improvements we're seeing. The "free" part can be removed because we're
using BN_CTX. The "clear" part OTOH can be removed because BN_CTX
destruction automatically performs this task, so performing it inside
functions that may be called repeatedly is wasteful. This is currently safe
within openssl due to the fact that BN_CTX objects are never created for
longer than a single high-level operation. However, that is only because
there's currently no mechanism in openssl for thread-local storage. Beyond
that, this might be an issue for applications using the bignum API directly
and caching their own BN_CTX objects. The solution is to introduce a flag
to BN_CTX_start() that allows its variables to be automatically sanitised
on release during BN_CTX_end(). This way any higher-level function (and
perhaps the application) can specify this flag in its own
BN_CTX_start()/BN_CTX_end() pair, and this will cause inner-loop functions
specifying the flag to be ignored so that sanitisation is handled only once
back out at the higher level. I will be implementing this in the near
future.
Geoff Thorpe [Thu, 25 Mar 2004 04:16:14 +0000 (04:16 +0000)]
Replace the BN_CTX implementation with my current work. I'm leaving the
little TODO list in there as well as the debugging code (only enabled if
BN_CTX_DEBUG is defined).
I'd appreciate as much review and testing as can be spared for this. I'll
commit some changes to other parts of the bignum code shortly to make
better use of this implementation (no more fixed size limitations). Note
also that under identical optimisations, I'm seeing a noticable speed
increase over openssl-0.9.7 - so any feedback to confirm/deny this on other
systems would also be most welcome.