Richard Levitte [Wed, 10 Feb 2016 21:33:44 +0000 (22:33 +0100)]
Rework BIO_ADDRINFO_protocol() to return correct values
As noted already, some platforms don't fill in ai_protocol as
expected. To circumvent that, we have BIO_ADDRINFO_protocol() to
compute a sensible answer in that case.
Matt Caswell [Wed, 10 Feb 2016 19:41:27 +0000 (19:41 +0000)]
Fix errstr error code parsing
Error codes are printed in hex, and previous OpenSSL versions expected
the error codes to be provided to errstr in hex. In 1.1.0, for some
reason, it was expecting them to be decimal.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Richard Levitte [Wed, 10 Feb 2016 22:27:28 +0000 (23:27 +0100)]
After auto init, check that the deprecated functions exist before using
The functions that have been deprecated by the auto init changes are
now guarded with deprecation checks, so it's fairly easy to see if
they can be used.
In test/dtlsv1listentest, we simply remove all init and cleanup code,
as they are call automatically when needed.
Richard Levitte [Wed, 10 Feb 2016 19:23:40 +0000 (20:23 +0100)]
Don't assert protocol equality
It seems that some platforms' getaddrinfo don't fill in the
ai_protocol field properly. On those, the assertion
'protocol == BIO_ADDRINFO_protocol(res)' will fail. Best to remove
it.
Richard Levitte [Wed, 10 Feb 2016 18:09:05 +0000 (19:09 +0100)]
Quote the CFLAG in Unixly Makefiles, for buildinf.h
Because the command line definitions of OPENSSLDIR and ENGINESDIR
contain quotes, we need a variant of CFLAG where backslashes and
quotes are escaped when we produce buildinf.h
Matt Caswell [Wed, 10 Feb 2016 15:16:06 +0000 (15:16 +0000)]
Attempt to log an error if init failed
If init failed we'd like to set an error code to indicate that. But if
init failed then when the error system tries to load its strings its going
to fail again. We could get into an infinite loop. Therefore we just set
a single error the first time around. After that no error is set.
Matt Caswell [Wed, 10 Feb 2016 16:46:33 +0000 (16:46 +0000)]
Variable was declared static when it shouldn't be
The local variable tmp was declared static when it shouldn't be. This
is in the no-threads implementation, and it was immediately initialised
to something else on every invokation of the function so it doesn't break
anything...but still shouldn't be there.
Richard Levitte [Sat, 30 Jan 2016 06:15:30 +0000 (07:15 +0100)]
clean away old VMS cruft
The old building scripts get removed, they are hopelessly gone in bit
rot by now.
Also remove the old symbol hacks. They were needed needed to shorten
some names to 31 characters, and to resolve other symbol clashes.
Because we now compile with /NAMES=(AS_IS,SHORTENED), this is no
longer required.
Richard Levitte [Sat, 30 Jan 2016 06:14:58 +0000 (07:14 +0100)]
unified build scheme: add a "unified" template for VMS descrip.mms
As part of this, change util/mkdef.pl to stop adding libraries to
depend on in its output. mkdef.pl should ONLY output a symbol
vector.
Because symbol names can't be longer than 31 characters, we use the
compiler to shorten those that are longer down to 23 characters plus
an 8 character CRC. To make sure users of our header files will pick
up on that automatically, add the DEC C supported extra headers files
__decc_include_prologue.h and __decc_include_epilogue.h.
Furthermore, we add a config.com, so VMS people can configure just as
comfortably as any Unix folks, thusly:
Richard Levitte [Sat, 30 Jan 2016 04:45:29 +0000 (05:45 +0100)]
unified build scheme: add the tweaks to build on Cygwin & Mingw
Cygwin and Mingw name their libraries a bit differently from the rest
of the POSIXly universe, we need to adapt to that.
In Makefile.tmpl, it means that some hunks will only be output
conditionally.
This also means that shared_extension for the Cygwin and Mingw
configurations in Configurations/10-main.conf are changing from .dll.a
to .dll. Makefile.shared does a fine job without having them
specified, and it's much easier to work with tucking an extra .a at
the end of files in the installation recipes than any amount of name
rewrites, especially with the support of the SHARED_NAME in the top
build.info.
Richard Levitte [Sat, 30 Jan 2016 02:21:39 +0000 (03:21 +0100)]
unified build scheme: give util/dofile.pl the possibility to output selectively
Under certain conditions, one might not want to output certain
sections of a template file. This adds the functions output_off() and
output_on(), reachable inside the templates. And example usage in a
Makefile template could be this:
@ : {- output_off() if $config{no_shared}; "" -}
... lines dealing with shared libraries
@ : {- output_on() -}
Richard Levitte [Sat, 30 Jan 2016 01:57:19 +0000 (02:57 +0100)]
Configure et al: move the installation directory logic to Makefiles
The logic to figure out the combinations of --prefix and --openssldir
has stayed in Configure so far, with Unix paths as defaults.
However, since we're making Configure increasingly platform agnostic,
these defaults need to change and adapt to the platform, along with
the logic to combine them.
The easiest to provide for this is to move the logic and the defaults
away from Configure and into the build files.
This also means that the definition of the macros ENGINESDIR and
OPENSSLDIR move away from include/openssl/opensslconf.h and into the
build files.
Richard Levitte [Sat, 30 Jan 2016 01:17:05 +0000 (02:17 +0100)]
Configure et al: treat C defines separately
With some compilers, C macros are defined differently on the command
line than on Unix. It could be that the flad to define them isn't -D,
it could also be that they need to be grouped together and not be mixed
in with the other compiler flags (that's how it's done on VMS, for
example).
On Unix family platform configurations, we can continue to have macro
definitions mixed in with the rest of the flags, so the changes in
Configurations/*.conf are kept to an absolute minimum.
Matt Caswell [Wed, 10 Feb 2016 09:47:51 +0000 (09:47 +0000)]
If we've not been inited don't deinit
If you call an explicit deinit when we've not been inited then a seg
fault can occur. We should check that we've been inited before attempting
to deinit.
Richard Levitte [Wed, 10 Feb 2016 04:12:52 +0000 (05:12 +0100)]
VMS getnameinfo() seems to have a bug with returned service string
It seems like it gives back gibberish. If we asked for a numeric
service, it's easy to check for a digit in the first position, and
if there isn't any, rewrite it using older methods.
Richard Levitte [Wed, 10 Feb 2016 01:18:38 +0000 (02:18 +0100)]
Simplify the specification of include dirs in the build dir
In build.info files, make the include directory in the build directory
absolute, or Configure will think it should be added to the source
directory top. Configure will turn it into a relative path if
possible.
Richard Levitte [Wed, 10 Feb 2016 01:00:37 +0000 (02:00 +0100)]
Make the processing of build.info files more aware of absolute dirs
There were cases where some input was absolute, and concatenating it
to the diretory to the source or build top could fail spectacularly.
Let's check the input first to see if it's absolute.
And while we're on the subject of checking if a file or dir spec is
absolute using file_name_is_absolute() has its own quirks on VMS,
where a logical name is considered absolute under most circumstances.
This is perfectly correct from a VMS point of view, but when parsing
the build.info files, we want single word file or directory names to
only be checked syntactically. A function isabsolute() that does the
right thing is the solution.
Matt Caswell [Tue, 9 Feb 2016 23:09:44 +0000 (23:09 +0000)]
NULL the thread_local_inits_st pointer after use
After the final use of the thread_local_inits_st we should ensure it is
set to NULL, just in case OPENSSL_INIT_thread_stop gets called again and
it tries to use garbage.
Matt Caswell [Tue, 9 Feb 2016 22:09:56 +0000 (22:09 +0000)]
Stop library before checking for mem leaks
With the new init framework resources aren't released until the process
exits. This means checking for mem leaks before that point finds a lot of
things! We should explicitly close down the library if we're checking for
mem leaks.
A J Mohan Rao [Tue, 9 Feb 2016 15:55:42 +0000 (10:55 -0500)]
GH646: Update help for s_server command.
* added missing help option messages
* ecdh_single option is removed as it is a no-op and not an option
supported in earlier versions
* ssl_ctx_security_debug() was invoked before ctx check for NULL
* trusted_first option can be removed, as it is always enabled in 1.1.
But not removed the option, require confirmation.
Signed-off-by: Rich Salz <rsalz@openssl.org> Reviewed-by: Matt Caswell <matt@openssl.org>
Matt Caswell [Tue, 9 Feb 2016 10:17:59 +0000 (10:17 +0000)]
Provide documentation for auto-init/auto-deinit
Provide some man pages for auto-init/deinit. Also update the INSTALL
documentation for information on the new Configure options implemented as
part of this.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Matt Caswell [Tue, 9 Feb 2016 11:26:14 +0000 (11:26 +0000)]
Provide framework for auto initialise/deinitialise of the library
This commit provides the basis and core code for an auto initialisation
and deinitialisation framework for libcrypto and libssl. The intention is
to remove the need (in many circumstances) to call explicit initialise and
deinitialise functions. Explicit initialisation will still be an option,
and if non-default initialisation is needed then it will be required.
Similarly for de-initialisation (although this will be a lot easier since
it will bring all de-initialisation into a single function).
Reviewed-by: Richard Levitte <levitte@openssl.org>
Richard Levitte [Tue, 9 Feb 2016 09:15:13 +0000 (10:15 +0100)]
Use rel2abs() on VMS, rather than realpath()
It seems realpath() is quite buggy on VMS, or will at least give quite
surprising results. On the other hand, realpath() is the better on
Unix to clean out clutter like foo/../bar on Unix.
So we make out own function to get the absolute directory for a given
input, and use rel2abs() or realpath() depending on the platform
Configure runs on.
Issue reported by Steven M. Schweda <sms@antinode.info>
Richard Levitte [Sat, 30 Jan 2016 00:05:33 +0000 (01:05 +0100)]
unified build scheme: adjust test framework for out of source build tree
To be able to run tests when we've built in a directory other than
the source tree, the testing framework needs a few adjustments.
test/testlib/OpenSSL/Test.pm needs to know where it can find
shlib_wrap.sh, and a number of other tests need to be told a different
place to find engines than what they may be able to figure out on
their own. Relying to $TOP is not enough, $SRCTOP and $BLDTOP can be
used as an alternative.
As part of this change, top_file and top_dir are removed and
srctop_file, bldtop_file, srctop_dir and bldtop_dir take their place.
Richard Levitte [Mon, 8 Feb 2016 04:11:25 +0000 (05:11 +0100)]
unified build scheme: add a design document
This documents describes the three steps from build.info files via the
%unified_info database to the build-file templates, along with some
examples showing how the data gets processed along the way.
Richard Levitte [Fri, 29 Jan 2016 23:57:33 +0000 (00:57 +0100)]
unified build scheme: add and document the "unified" driving engine
common.tmpl will be used together with the template build file, and is
the engine that connects the information gathered from all the
build.info files with making the build file itself.
This file expects there to be a template section in the build file
template that defines a number perl functions designed to return
strings with appropriate lines for the build system at hand. The
exact functions, what they can expect as arguments and what output
they're expected to produce is documented in Configurations/README.
Viktor Dukhovni [Mon, 8 Feb 2016 00:07:57 +0000 (19:07 -0500)]
Suppress DANE TLSA reflection when verification fails
As documented both SSL_get0_dane_authority() and SSL_get0_dane_tlsa()
are expected to return a negative match depth and nothing else when
verification fails. However, this only happened when verification
failed during chain construction. Errors in verification of the
constructed chain did not have the intended effect on these functions.
This commit updates the functions to check for verify_result ==
X509_V_OK, and no longer erases any accumulated match information
when chain construction fails. Sophisticated developers can, with
care, use SSL_set_verify_result(ssl, X509_V_OK) to "peek" at TLSA
info even when verification fail. They must of course first check
and save the real error, and restore the original error as quickly
as possible. Hiding by default seems to be the safer interface.
Introduced X509_V_ERR_DANE_NO_MATCH code to signal failure to find
matching TLSA records. Previously reported via X509_V_ERR_CERT_UNTRUSTED.
This also changes the "-brief" output from s_client to include
verification results and TLSA match information.
Mentioned session resumption in code example in SSL_CTX_dane_enable(3).
Also mentioned that depths returned are relative to the verified chain
which is now available via SSL_get0_verified_chain(3).
Added a few more test-cases to danetest, that exercise the new
code.
Resolved thread safety issue in use of static buffer in
X509_verify_cert_error_string().
Fixed long-stating issue in apps/s_cb.c which always sets verify_error
to either X509_V_OK or "chain to long", code elsewhere (e.g.
s_time.c), seems to expect the actual error. [ The new chain
construction code is expected to correctly generate "chain
too long" errors, so at some point we need to drop the
work-arounds, once SSL_set_verify_depth() is also fixed to
propagate the depth to X509_STORE_CTX reliably. ]