essentially overwrites itself with the new ENGINE, with the exception of
reference counts, ex_data structures, and other 'admin' elements. However
if the new ENGINE doesn't populate certain elements, there's the risk of
the "dynamic" ENGINE's elements showing through - the "cmd_defns" were just
one of the possibilities. This implements a more comprehensive cleanup.
empty set. This prevents engines that do not set the command
definitions themselves to inherit the ones from "dynamic", which would
otherwise be very confusing.
sooner and the programs get built against the shared libraries.
This requires a bit more work. Things like -rpath and the possibility
to still link the programs statically should be included. Some
cleanup is also needed. This will be worked on.
libdes (which is still used out there) or other des implementations,
the OpenSSL DES functions are renamed to begin with DES_ instead of
des_. Compatibility routines are provided and declared by including
openssl/des_old.h. Those declarations are the same as were in des.h
when the OpenSSL project started, which is exactly how libdes looked
at that time, and hopefully still looks today.
The compatibility functions will be removed in some future release, at
the latest in version 1.0.
such cases, a flush should *not* attempt to finalise the encoding, as
the EVP_ENCODE_CTX structure will only be filled with garbage. For
the same reason, do the same check when a wpending is performed.
not implemented. (Bug reported by Martin Szotkowski)
This also changes the non-"_ex" versions to defer directly to
EVP_CipherInit_ex() rather than EVP_CipherInit() to avoid an unecessary
level of indirection.
it to be defined on all platforms whether or not it is of any practical
use on them. This also resolves linker problems on "special" platforms,
such as win32.
of the stack, and the (void *) type used in the underlying sk_***
functions. However, declaring a STACK_OF(type) where type is a *function*
type implicitly involves casts between function pointers and data pointers.
That's a no-no. This changes the ENGINE_CLEANUP handling to use a regular
data type in the stack.
ENGINE redevelopment. The idea had been that "-1" could be used as a
special "ask me later" 'nid' rather than specifying supported cipher and
digest 'nid's up front. However the idea turned out to be pretty broken.
testing. Because of the recent changes (see crypto/engine/README), the
"openssl" ENGINE is no longer needed nor is it loaded automatically or by
ENGINE_load_builtin_engines(). So a explicit ENGINE_load_openssl() call is
required by applications or a modification to eng_all.c before this ENGINE
will be used. This change will send output to stderr as/when its
implementations are used.
See crypto/engine/README for details.
- it also removes openbsd_hw.c from the build (that functionality is
going to be available in the openbsd ENGINE in a upcoming commit)
- evp_test has had the extra initialisation added so it will use (if
possible) any ENGINEs supporting the algorithms required.
ENGINE surgery. DH, DSA, RAND, and RSA now use *both* "method" and ENGINE
pointers to manage their hooking with ENGINE. Previously their use of
"method" pointers was replaced by use of ENGINE references. See
crypto/engine/README for details.
Also, remove the ENGINE iterations from evp_test - even when the
cipher/digest code is committed in, this functionality would require a
different set of API calls.
ENGINE_TABLE-based stuff - as described in crypto/engine/README.
Associated miscellaneous changes;
- the previous cipher/digest hooks that hardwired directly to EVP's
OBJ_NAME-based storage have been backed out. New cipher/digest support
has been constructed and will be committed shortly.
- each implementation defines its own ENGINE_load_<name> function now.
- the "openssl" ENGINE isn't needed or loaded any more.
- core (not algorithm or class specific) ENGINE code has been split into
multiple files to increase readability and decrease linker bloat.
- ENGINE_cpy() has been removed as it wasn't really a good idea in the
first place and now, because of registration issues, can't be
meaningfully defined any more.
- BN_MOD_EXP[_CRT] support is removed as per the README.
- a bug in enginetest.c has been fixed.
NB: This commit almost certainly breaks compilation until subsequent
changes are committed.
digest support, are on their way. Rather than having gigantic commit log
messages and/or CHANGES entries, this change to the README will serve as an
outline of what it all is and how it all works.
in "types.h" so that very few headers will need to include engine.h,
generally only C files using API functions will need it (reducing
the header dependencies quite a lot).
distinction (which does not work well because if CRYPTO_MDEBUG is
defined at library compile time, it is not necessarily defined at
application compile time; and memory debugging now can be reconfigured
at run-time anyway). To get the intended semantics, we could just use
the EVP_DigestInit_dbg unconditionally (which uses the caller's
__FILE__ and __LINE__ for memory leak debugging), but this would make
memory debugging inconsistent. Instead, callers can use
CRYPTO_push_info() to track down memory leaks.
distinction (which does not work well because if CRYPTO_MDEBUG is
defined at library compile time, it is not necessarily defined at
application compile time; and memory debugging now can be reconfigured
at run-time anyway). To get the intended semantics, we could just use
the EVP_DigestInit_dbg unconditionally (which uses the caller's
__FILE__ and __LINE__ for memory leak debugging), but this would make
memory debugging inconsistent. Instead, callers can use
CRYPTO_push_info() to track down memory leaks.
Also fix indentation, and add OpenSSL copyright.
CRYPTO_set_mem_debug_options() instead of CRYPTO_dbg_set_options(),
which is the default implementation of the former and should usually
not be directly used by applications (at least if we assume that the
options accepted by the default implementation will also be meaningful
to any other implementations).
Also fix apps/openssl.c and ssl/ssltest such that environment variable
setting 'OPENSSL_DEBUG_MEMORY=off' actively disables the compiled-in
library defaults (i.e. such that CRYPTO_MDEBUG is ignored in this
case).
(Some platforms need _XOPEN_SOURCE and _XOPEN_SOURCE_EXTENDED to get
the declaration, but on other platforms _XOPEN_SOURCE disables
the strdup declaration in <string.h>.)
In err.c, flags int_error_hash_set and int_thread_hash_set
appear superfluous since we can just as well initialize
int_error_hash and int_thread_hash to NULL.
Change some of the err.c formatting to conform with the rest of
OpenSSL.
error strings - the destroy handler functions unload the error strings so
any pending error state referring to them will not attempt to reference
them after the ENGINE has been destroyed.
being enabled or disabled (respectively) for operation. Additionally, each
ENGINE has a constructor function where it can do more 'structural' level
intialisations such as loading error strings, creating "ex_data" indices,
etc. This change introduces a handler function that gives an ENGINE a
corresponding opportunity to cleanup when the ENGINE is being destroyed. It
also adds the "get/set" API functions that control this "destroy" handler
function in an ENGINE.
declaration and implementation had not. So a recent update recreated the
original definition in libeay.num ... this corrects it and changes the "dh"
code to the "up_ref" variant.
defined.
(Preprocessor symbols such as _POSIX_C_SOURCE or _XOPEN_SOURCE are
supposed to disable anything not allowed by the respective
specification; I'm not sure why 'strdup' would be considered
an outlaw though.)
locking callbacks to pass to the loaded library (in addition to the
existing mem, ex_data, and err callbacks). Also change the default
implementation of the "bind_engine" function to apply those callbacks, ie.
the IMPLEMENT_DYNAMIC_BIND_FN macro.
declare their own error strings so that they can be more easily compiled as
external shared-libraries if desired. Also, each implementation has been
given canonical "dynamic" support at the base of each file and is only
built if the ENGINE_DYNAMIC_SUPPORT symbol is defined.
Also, use "void" prototypes rather than empty prototypes in engine_int.h.
This does not yet;
(i) remove error strings when unloading,
(ii) remove the redundant ENGINE_R_*** codes (though ENGINE_F_*** codes
have gone), or
(iii) provide any instructions on how to build shared-library ENGINEs or
use them.
All are on their way.
implementations to be loaded from self-contained shared-libraries. It also
provides (in engine.h) definitions and macros to help implement a
self-contained ENGINE. Version control is handled in a way whereby the
loader or loadee can veto the load depending on any objections it has with
each other's declared interface level. The way this is currently
implemented assumes a veto will only take place when one side notices the
other's interface level is too *old*. If the other side is newer, it should
be assumed the newer version knows better whether to veto the load or not.
Version checking (like other "dynamic" settings) can be controlled using
the "dynamic" ENGINE's control commands. Also, the semantics for the
loading allow a shared-library ENGINE implementation to handle differing
interface levels on the fly (eg. loading secondary shared-libraries
depending on the versions required).
Code will be added soon to the existing ENGINEs to illustrate how they can
be built as external libraries rather than building statically into
libcrypto.
NB: Applications wanting to support "dynamic"-loadable ENGINEs will need to
add support for ENGINE "control commands". See apps/engine.c for an example
of this, and use "apps/openssl engine -vvvv" to test or experiment.
Henson). Also, reverse a previous change that used an implicit function
pointer cast rather than an explicit data pointer cast in the STACK cleanup
code.
See the commit log message for that for more information.
NB: X509_STORE_CTX's use of "ex_data" support was actually misimplemented
(initialisation by "memset" won't/can't/doesn't work). This fixes that but
requires that X509_STORE_CTX_init() be able to handle errors - so its
prototype has been changed to return 'int' rather than 'void'. All uses of
that function throughout the source code have been tracked down and
adjusted.
Currently, this change merely addresses where ex_data indexes are stored
and managed, and thus fixes the thread-safety issues that existed at that
level. "Class" code (eg. RSA, DSA, etc) no longer store their own STACKS
and per-class index counters - all such data is stored inside ex_data.c. So
rather than passing both STACK+counter to index-management ex_data
functions, a 'class_index' is instead passed to indicate the class (eg.
CRYPTO_EX_INDEX_RSA). New classes can be dynamically registered on-the-fly
and this is also thread-safe inside ex_data.c (though whether the caller
manages the return value in a thread-safe way is not addressed).
This does not change the "get/set" functions on individual "ex_data"
structures, and so thread-safety at that level isn't (yet) assured.
Likewise, the method of getting and storing per-class indexes has not
changed, so locking may still be required at the "caller" end but is
nonetheless thread-safe inside "ex_data"'s internal implementation.
Typically this occurs when code implements a new method of some kind and
stores its own per-class index in a global variable without locking the
setting and usage of that variable. If the code in question is likely to be
used in multiple threads, locking the setting and use of that index is
still up to the code in question. Possible fixes to this are being
sketched, but definitely require more major changes to the API itself than
this change undertakes.
The underlying implementation in ex_data.c has also been modularised so
that alternative "ex_data" implementations (that control all access to
state) can be plugged in. Eg. a loaded module can have its implementation
set to that of the application loaded it - the result being that
thread-safety and consistency of "ex_data" classes and indexes can be
maintained in the same place rather than the loaded module using its own
copy of ex_data support code and state.
Due to the centralisation of "state" with this change, cleanup of all
"ex_data" state can now be performed properly. Previously all allocation of
ex_data state was guaranteed to leak - and MemCheck_off() had been used to
avoid it flagging up the memory debugging. A new function has been added to
perfrom all this cleanup, CRYPTO_cleanup_all_ex_data(). The "openssl"
command(s) have been changed to use this cleanup, as have the relevant test
programs. External application code may want to do so too - failure to
cleanup will not induce more memory leaking than was the case before, but
the memory debugging is not tricked into hiding it any more so it may
"appear" where it previously did not.
error strings and a hash table storing per-thread error state) go via an
ERR_FNS function table. The first time an ERR operation occurs, the
implementation that will be used (from then on) is set to the internal
"defaults" implementation if it has not already been set. The actual LHASH
tables are only accessed by this implementation.
This is primarily for modules that can be loaded at run-time and bound into
an application (or a shared-library version of OpenSSL). If the module has
its own statically-linked copy of OpenSSL code - this mechanism allows it
to *not* create and use ERR information in its own linked "ERR" code, but
instead to use and interact with the state stored in the loader
(application or shared library). The loader calls ERR_get_implementation()
and the return value is what the module should use when calling its own
copy of ERR_set_implementation().
dependant code has to directly increment the "references" value of each
such structure using the corresponding lock. Apart from code duplication,
this provided no "REF_CHECK/REF_PRINT" checking and violated
encapsulation.
setting stack (actually, array) values in ex_data. So only increment the
global counters if the underlying CRYPTO_get_ex_new_index() call succeeds.
This change doesn't make "ex_data" right (see the comment at the head of
ex_data.c to know why), but at least makes the source code marginally less
frustrating.
setting stack (actually, array) values in ex_data. So only increment the
global counters if the underlying CRYPTO_get_ex_new_index() call succeeds.
This change doesn't make "ex_data" right (see the comment at the head of
ex_data.c to know why), but at least makes the source code marginally less
frustrating.
His comments are:
1) Changes all references for `True64' to be `Tru64', which is the correct
spelling for the OS name.
2) Makes `alpha-cc' be the same as `alpha164-cc', and adds an `alphaold-cc'
entry that is the same as the previous `alpha-cc'. The reason is that most
people these days are using the newer compiler, so it should be the default.
3) Adds a bit of commentary to Configure, regarding the name changes of
the OS over the years, so it's not so confusing to people that haven't been
with the OS for a while.
4) Adds an `alpha-cc-rpath' target (which is *not* selected automatically
by Configure under any circumstance) that builds an RPATH into the
shared libraries. This is explained in the comment in Configure. It's
very very useful for people that want it, and people that don't want it
just shouldn't choose that target.
5) Adds the `-pthread' flag as the best way to get POSIX thread support
from the newer compiler.
6) Updates the Makefile targets, so that when the `alpha164-cc', `alpha-cc',
or `alpha-cc-rpath' target is what Configure is set to use, it uses a Makefile
target that includes the `-msym' option when building the shared library.
This is a performance enhancement.
7) Updates `config' so that if it detects you're running version 4 or 5
of the OS, it automatically selects `alpha-cc', but uses `alphaold-cc'
for versions 1-3 of the OS.
8) Updates the comment in opensslv.h, fixing both the OS name typo and
adding a reference to IRIX 6.x, since the shared library semantics are
virtually identical there.
DES's keyschedules.
I know these two should be separate, and I'll back out the DES changes if they
are deemed to be an error.
Note that there is a memory leak lurking in SSL somewhere in this version.
HP-UX in common in ./config). Note that for the moment of this writing
none of 64-bit platforms pass bntest. I'm committing this anyway as it's
too frustrating to patch snapshots over and over while 0.9.6 is known to
work.
Split private key PEM and normal PEM handling. Private key
handling needs to link in stuff like PKCS#8.
Relocate the ASN1 *_dup() functions, to the relevant ASN1
modules using new macro IMPLEMENT_ASN1_DUP_FUNCTION. Previously
these were all in crypto/x509/x_all.c along with every ASN1
BIO/fp function which linked in *every* ASN1 function if
a single dup was used.
Move the authority key id ASN1 structure to a separate file.
This is used in the X509 routines and its previous location
linked in all the v3 extension code.
Also move ASN1_tag2bit to avoid linking in a_bytes.c which
is now largely obsolete.
So far under Linux stripped binary with single PEM_read_X509
is now 238K compared to 380K before these changes.
reduce linker bloat. For example the
single line:
PEM_read_X509()
results in a binary of around 400K in Linux!
This first step separates some of the PEM functions and
avoids linking in some PKCS#7 and PKCS#12 code.
does not contain more bytes than the RSA modulus 'n' - it does not check
that the input is strictly *less* than 'n'. Whether this should be the
case or not is open to debate - however, due to security problems with
returning miscalculated CRT results, the 'rsa_mod_exp' implementation in
rsa_eay.c now performs a public-key exponentiation to verify the CRT result
and in the event of an error will instead recalculate and return a non-CRT
(more expensive) mod_exp calculation. As the mod_exp of 'I' is equivalent
to the mod_exp of 'I mod n', and the verify result is automatically between
0 and n-1 inclusive, the verify only matches the input if 'I' was less than
'n', otherwise even a correct CRT calculation is only congruent to 'I' (ie.
they differ by a multiple of 'n'). Rather than rejecting correct
calculations and doing redundant and slower ones instead, this changes the
equality check in the verification code to a congruence check.
SSL according to RFC 2712. His comment is:
This is a patch to openssl-SNAP-20010702 to support Kerberized SSL
authentication. I'm expecting to have the full kssl-0.5 kit up on
sourceforge by the end of the week. The full kit includes patches
for mod-ssl, apache, and a few text clients. The sourceforge URL
is http://sourceforge.net/projects/kssl/ .
Thanks to a note from Simon Wilkinson I've replaced my KRB5 AP_REQ
message with a real KerberosWrapper struct. I think this is fully
RFC 2712 compliant now, including support for the optional
authenticator field. I also added openssl-style ASN.1 macros for
a few Kerberos structs; see crypto/krb5/ if you're interested.
Add new extension functions which work with NCONF.
Tidy up extension config routines and remove redundant code.
Fix NCONF_get_number().
Todo: more testing of apps to see they still work...
