Unified - adapt the generation of des assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/des/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of ec assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/ec/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of camellia assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/camellia/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of cast assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/cast/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of chacha assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/chacha/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Because crypto/build.info demands CFLAGS_Q, descrip.mms.tmpl must deliver

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of cpuid, uplink and buildinf to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of aes assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/aes/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Unified - adapt the generation of blowfish assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/bf/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Convert the dynlocks in e_chil to the new Thread API locks

Reviewed-by: Matt Caswell <matt@openssl.org>

Remove another lock from e_chil

Reviewed-by: Richard Levitte <levitte@openssl.org>

Always call ENGINE_cleanup() in de-init

Even if we haven't loaded an engine, we might have set up the
global_engine_lock, so we should still clean up.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move chil engine to the new thread api

Move the chil engine to use the new thread API. As I don't have access to
the hardware I can't test this :-(. I think its ok...

Reviewed-by: Richard Levitte <levitte@openssl.org>

Move engine library over to using the new thread API

Remove usage of CRYPTO_LOCK_ENGINE

Reviewed-by: Richard Levitte <levitte@openssl.org>

err_lcl.h is gone, don't pretend it's there

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>

Add missing semi

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>

Counter mixed signedness with a cast

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>

Fix names of the #define used for platform specific code

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Add empty line after local variables

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix error with wait set of fds for the select()

It also makes the call to select blocking to reduce CPU usage

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Use correct function ID in error path

This fixes "make update".

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

Move variable declaration to the start of the function

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

Ensure CRYPTO_mem_leaks is the last thing we do

CRYPTO_mem_leaks de-inits the library, so we must not do anything
interesting after we've used it!

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix memory leak in ssltest

The new Rand usage of Thread API exposed a bug in ssltest. ssltest "cheats"
and uses internal headers to directly call functions that normally you
wouldn't be able to do. This means that auto-init doesn't happen, and
therefore auto-deinit doesn't happen either, meaning that the new rand locks
don't get cleaned up properly.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert rand code to new threading API

Replace the CRYPTO_LOCK_RAND and CRYPTO_LOCK_RAND2 locks with new thread
API style locks.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Adapt unix Makefile template to 'no-makedepend'

This change is a bit more complex, as it involves several recipe
variants.

Also, remove the $(CROSS_COMPILE) prefix for the makedepend program.
When we use the program "makedepend", this doesn't serve anything,
and when we use the compiler, this value isn't even used.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Adapt descrip.mms.tmpl to 'no-makedepend'

VMS doesn't have "makedepend" anyway, so this is just a matter of using
the right qualifiers when 'makedepend' is enabled.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Add the configure option 'no-makedepend'

If no makedepend program or equaly capable compiler is present,
'makedepend' gets disabled automatically.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix SSL_CIPHER_get_auth_nid return

Copy/paste error between SSL_CIPHER_get_kx_nid() and
SSL_CIPHER_get_auth_nid(), wrong table was referenced

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

Fix a typo in dynamic_load()

Reviewed-by: Matt Caswell <matt@openssl.org>

Convert RSA blinding to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_SSL_* to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Remove the transfer of lock hooks from bind_engine

With the new threads API, this is no longer needed.

Reviewed-by: Matt Caswell <matt@openssl.org>

Engine API repair - memory management hooks

The Engine API lost the setting of memory management hooks in
bind_engine.  Here's putting that back.

EX_DATA and ERR functions need the same treatment.

Reviewed-by: Matt Caswell <matt@openssl.org>

Redo the Unix source code generator

For assembler, we want the final target to be foo.s (lowercase s).
However, the build.info may have lines like this (note upper case S):

    GENERATE[foo.S]=foo.pl

This indicates that foo.s (lowercase s) is still to be produced, but
that producing it will take an extra step via $(CC) -E.  Therefore,
the following variants (simplified for display) can be generated:

    GENERATE[foo.S]=foo.pl  =>  foo.s: foo.pl
                                 $(PERL) $foo.pl $@.S; \
                                 $(CC) $(CFLAGS) -E -P $@.S > $@ && \
                                 rm -f $@.S

    GENERATE[foo.s]=foo.pl  =>  foo.s: foo.pl
                                 $(PERL) $foo.pl $@

    GENERATE[foo.S]=foo.m4  =>  foo.s: foo.m4
                                 m4 -B 8192 $foo.m4 > $@.S; \
                                 $(CC) $(CFLAGS) -E -P $@.S > $@ && \
                                 rm -f $@.S

    GENERATE[foo.s]=foo.m4  =>  foo.s: foo.m4
                                 m4 -B 8192 $foo.m4 > $@

Reviewed-by: Andy Polyakov <appro@openssl.org>

Fix error in ssltest

Compiling ssltest with some compilers using --strict-warnings results in
complaints about an unused result.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>

Adapt e_capi to the DSA_SIG_get0() API

Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert mem_dbg and mem_sec to the new Thread API

Use new Thread API style locks, and thread local storage for mem_dbg

Reviewed-by: Rich Salz <rsalz@openssl.org>

OpenSSLDie --> OPENSSL_die

Also removed a bunch of unused define's from e_os.h

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove the -n tar flag from osx dist creation

With the unified build scheme, tar MUST recurse, or the tarball will
be empty.

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>

Remove some old ms/* files

Reviewed-by: Richard Levitte <levitte@openssl.org>

Include e_os.h from ec_lcl.h

The EC code recently started using REF_PRINT_COUNT and REF_ASSERT_ISNT.
Those are defined in e_os.h.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix cert leaks in s_server

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

Reuse strndup(), simplify code

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

make update

Reviewed-by: Rich Salz <rsalz@openssl.org>

Make DSA_SIG opaque.

This adds a new accessor function DSA_SIG_get0.
The customisation of DSA_SIG structure initialisation has been removed this
means that the 'r' and 's' components are automatically allocated when
DSA_SIG_new() is called. Update documentation.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Fix travis builds

Travis doesn't seem to know about 'expr'

Reviewed-by: Emilia Käsper <emilia@openssl.org>

make update

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_X509_* to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_UI to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_EC_* to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_EVP_PKEY to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_BIO to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Only enable ccache if it's available

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Use ccache for the unified builds

This may speed up our builds considerably

Reviewed-by: Emilia Käsper <emilia@openssl.org>

For unified builds, make a separate build directory and build there

Reviewed-by: Emilia Käsper <emilia@openssl.org>

SPARCv9 assembly pack: unify build rules and argument handling.

Make all scripts produce .S, make interpretation of $(CFLAGS)
pre-processor's responsibility, start accepting $(PERLASM_SCHEME).
[$(PERLASM_SCHEME) is redundant in this case, because there are
no deviataions between Solaris and Linux assemblers. This is
purely to unify .pl->.S handling across all targets.]

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix build; ssltest

Reviewed-by: Richard Levitte <levitte@openssl.org>

Tweak some more information in INSTALL

The summary on how to add configs and how Makefile et al wasn't quite
correct any more.

Reviewed-by: Rich Salz <rsalz@openssl.org>

GH787: Fix ALPN

* Perform ALPN after the SNI callback; the SSL_CTX may change due to
  that processing
* Add flags to indicate that we actually sent ALPN, to properly error
  out if unexpectedly received.
* clean up ssl3_free() no need to explicitly clear when doing memset
* document ALPN functions

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Add an SSL_has_pending() function

This is similar to SSL_pending() but just returns a 1 if there is data
pending in the internal OpenSSL buffers or 0 otherwise (as opposed to
SSL_pending() which returns the number of bytes available). Unlike
SSL_pending() this will work even if "read_ahead" is set (which is the
case if you are using read pipelining, or if you are doing DTLS). A 1
return value means that we have unprocessed data. It does *not* necessarily
indicate that there will be application data returned from a call to
SSL_read(). The unprocessed data may not be application data or there
could be errors when we attempt to parse the records.

Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Add an ability to set the SSL read buffer size

This capability is required for read pipelining. We will only read in as
many records as will fit in the read buffer (and the network can provide
in one go). The bigger the buffer the more records we can process in
parallel.

Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Add cipher query functions

Add functions to determine authentication, key-exchange, FIPS and AEAD.

Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert ERR_STRING_DATA to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert ERR_STATE to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_EX_DATA to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_DSO to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_{DH,DSA,RSA} to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Convert CRYPTO_LOCK_GET*BYNAME to new multi-threading API

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

GH815: The ChaCha20/Poly1305 codepoints are official

CCA8, CCA9, CCAA, CCAB, CCAC, CCAD, and CCAE are now present in
https://www.iana.org/assignments/tls-parameters/tls-parameters.xhtml
so remove the "as per draft-ietf-tls-chacha20-poly1305-03" note
accordingly.

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

GH787: Fix ALPN

* Perform ALPN after the SNI callback; the SSL_CTX may change due to
  that processing
* Add flags to indicate that we actually sent ALPN, to properly error
  out if unexpectedly received.
* clean up ssl3_free() no need to explicitly clear when doing memset
* document ALPN functions

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

Make mk1mf recognise the --classic flag

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix configurations such as 'dist' and tar building

For config targets such as 'dist', which doesn't have a BASE template,
we still need to have a default build scheme.

Additionally, the unified Makefile template's target 'tar' wasn't
quite as flexible as the unixmake one.

Finally, .travis-create-release.sh can be somewhat simplified now that
it builds with the unified build scheme.

Reviewed-by: Matt Caswell <matt@openssl.org>

We've switch to unified build scheme by default, reflect it in travis

We want to preserve a couple of classic builds still, the quickest is
to change all --unified to --classic

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix engine/asm/e_padlock-x86.pl for newer semantics

Reviewed-by: Andy Polyakov <appro@openssl.org>

Change the INSTALL documentation for unified builds

Because of the unified scheme, building on different platforms is very
similar.  We currently have Unix and OpenVMS on the unified scheme,
which means that a separate INSTALL.VMS is no longer needed.

Reviewed-by: Matt Caswell <matt@openssl.org>

Make unified builds the default on Unix

Reviewed-by: Matt Caswell <matt@openssl.org>

Add the Configure option --classic, to fall back on classic build schemes

Reviewed-by: Matt Caswell <matt@openssl.org>

Revert "unified build scheme: Try to nudge users to try the "unified" build"

This reverts commit 242ffb05a2e4aa3fc7ffc131037e077b7e242189.

Reviewed-by: Matt Caswell <matt@openssl.org>

Fix building without multiblock support

Not all platforms support multiblock. Building without it fails prior to
this fix.

RT#4396

Reviewed-by: Richard Levitte <levitte@openssl.org>

Don't free NCONF obtained values

Bug reported by Michel Sales.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Remove kinv/r fields from DSA structure.

The kinv/r fields in the DSA structure are not used by OpenSSL internally
and should not be used in general.

Reviewed-by: Rich Salz <rsalz@openssl.org>

GH812: Fix for no-ui build on Windows

Add UI to known algorithms in mkdef.pl

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>

AppVeyor: Only use the latest VS version

Reviewed-by: Rich Salz <rsalz@openssl.org>
GH: #811

Rename the numpipes argument to ssl3_enc/tls1_enc

The numpipes argument to ssl3_enc/tls1_enc is actually the number of
records passed in the array. To make this clearer rename the argument to
|n_recs|.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Rename a function

Rename the have_whole_app_data_record_waiting() function to include the
ssl3_record prefix...and make it a bit shorter.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Refactor dasync cipher implementations to improve code reuse

Move out most of the boiler plate code that is common between aes128-cbc
and aes128-cbc-hmac-sha1 into helper functions to improve code reuse.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Move the _hidden_* static variables in dasync to be constructed in bind

The _hidden_* variables were being created on-the-fly. It is better to
create them once up front during bind to avoid any potential race
conditions.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Fix typo in SSL_pending docs

Reviewed-by: Tim Hudson <tjh@openssl.org>

Rename EVP_CIPHER_CTX_cipher_data to EVP_CIPHER_CTX_get_cipher_data

We had the function EVP_CIPHER_CTX_cipher_data which is newly added for
1.1.0. As we now also need an EVP_CIPHER_CTX_set_cipher_data it makes
more sense for the former to be called EVP_CIPHER_CTX_get_cipher_data.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Fix s_server/s_client handling of the split_send_frag argument

Ensure that a value of 0 is correctly handled for the split_send_frag
argument.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Add documentation for the EVP_CIPHER_CTX_cipher_data functions

The new pipeline code added a new function
EVP_CIPHER_CTX_set_cipher_data(). Add documentation for this and the
existing EVP_CIPHER_CTX_cipher_data() function.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Remove the wrec record layer field

We used to use the wrec field in the record layer for keeping track of the
current record that we are writing out. As part of the pipelining changes
this has been moved to stack allocated variables to do the same thing,
therefore the field is no longer needed.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Update a comment

Update a comment that was out of date due to the pipelining changes

Reviewed-by: Tim Hudson <tjh@openssl.org>

Add documentation for new s_server/s_client options

Document the new split_send_frag, max_pipelines and read_buf options.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Add documentation for SSL_has_pending()

A previous commit added the SSL_has_pending() function which provides a
method for knowing whether OpenSSL has buffered, but as yet unprocessed
record data.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Add pipelining documentation

Add some documentation for all of the SSL/SSL_CTX functions/ctrls for
conrolling read and write pipelining.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Fix erroneous fall thgrough in switch statement

Fix an erroenous fall through when setting the max_pipelines value.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Ensure s_client and s_server work when read_ahead is set

Previously s_client and s_server relied on using SSL_pending() which does
not take into account read_ahead. For read pipelining to work, read_ahead
gets set automatically. Therefore s_client and s_server have been
converted to use SSL_has_pending() instead.

Reviewed-by: Tim Hudson <tjh@openssl.org>

Add an SSL_has_pending() function

This is similar to SSL_pending() but just returns a 1 if there is data
pending in the internal OpenSSL buffers or 0 otherwise (as opposed to
SSL_pending() which returns the number of bytes available). Unlike
SSL_pending() this will work even if "read_ahead" is set (which is the
case if you are using read pipelining, or if you are doing DTLS). A 1
return value means that we have unprocessed data. It does *not* necessarily
indicate that there will be application data returned from a call to
SSL_read(). The unprocessed data may not be application data or there
could be errors when we attempt to parse the records.

Reviewed-by: Tim Hudson <tjh@openssl.org>