Make BIO opaque

Move the the BIO_METHOD and BIO structures into internal header files,
provide appropriate accessor methods and update all internal code to use
the new accessors where appropriate.

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

Add functions for creating BIO_METHODs

BIO_METHODs are soon to be opaque so we need to have functions available
to set them up.

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

Fix a big leak when using stack-allocated BIO items.

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

poly1305/asm/poly1305-x86.pl: don't loose 59-th bit.

RT#4439

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

crypto/seed: add small-footprint path.

The space saving is >5x on x86_64 at ~40% performance penalty.

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

Windows build system: fix 32-bit appveyor build.

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

Build system: VC-WIN64I fixups.

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

When using no-dtls create OPENSSL_NO_DTLS

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
MR: #2452

Add session reuse tests.

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
MR: #2452

Resolve DTLS cookie and version before session resumption.

Session resumption involves a version check, so version negotiation must
happen first. Currently, the DTLS implementation cannot do session
resumption in DTLS 1.0 because the ssl_version check always checks
against 1.2.

Switching the order also removes the need to fixup ssl_version in DTLS
version negotiation.

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
RT: #4392, MR: #2452

It's called SSL_session_reused(), not SSL_session_resumed()

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
MR: #2452

Allow different protocol version when trying to reuse a session

We now send the highest supported version by the client, even if the session
uses an older version.

This fixes 2 problems:
- When you try to reuse a session but the other side doesn't reuse it and
  uses a different protocol version the connection will fail.
- When you're trying to reuse a session with an old version you might be
  stuck trying to reuse the old version while both sides support a newer
  version

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
GH: #852, MR: #2452

Math::BigInt does floored divs, BN_div does truncated div, compensate

According to documentation, perl's Math::BigInt does floored division,
i.e. the bdiv function does 1 / -4 = -1.  OpenSSL's BN_div, as well as
bc, do truncated division, i.e. 1 / -4 = 0.

We need to compensate for that difference in test/recipes/bc.pl to
make sure to verify the bntest results under its own conditions, by
dividing the absolute values of the given numbers and fixup the
result's negativity afterwards.

Closes RT#4485

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

RT4476: Fix some cast-alignment warnings

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

VMS: add a note about DECC$* logical names

These logical names are used to make the C RTL change certain
behaviors, sometimes to make it act more like Unix.  While they can
make life easier in some cases, they can be disruptive as well.  When
building and testing OpenSSL, the latter is the case, so we ask people
to avoid using them.

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

Revert part of applink/Borland commit

This allows developer to glue DLL built with VC into their application
compiled with Borland C.

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

More Borland removal.

And thanks to Miod Vallat for the nudge about ERR_PACK :)

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

VMS: $? might be huge enough to break line length limit, so don't print it

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

VMS: install openssl.conf in OPENSSLDIR, not INSTALLTOP

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

VMS: have mms ignore creation of already existing dirs

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

VMS: Rethink the staging directory

On Windows and Unix, the staging directory $(DESTDIR) can simply be
prepended to the installation directory.  An attempt was made to do
something similar on VMS, but that ended up being a half measure
solution.  Instead of that, simply use the staging directory as a
prefix under which [.OPENSSL-INSTALL] and [.OPENSSL-COMMON] will hold
the two directory trees that should end up in the directories
indicated by --prefix and --openssldir, and finish the installation
with appropriate instructions on what to do next.

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

Misc fixes in VMS/openssl_startup.com.in

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

Adjust some default installation directories

- on VMS, SYS$COMMON:[SSL] is already used as installation directory
  by HP SSL, so we make our default for --openssldir
  SYS$COMMON:[OPENSSL-COMMON] instead.
- Updated notes on default installation dirs fir Unix and Windows

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

VMS: update the properties of symbol search

In this OpenSSL version, we deliver engines with lower case symbol
names.  The DSO symbol finder must be updated to allow for mixed case
symbols or it won't fine them.

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

VMS: compensate for command line length limits with a logical name

Sometimes, you might end up with a rather long compile line due to
excessively long /INCLUDE directories.  Compensate for it by making
a temporary logical name with them and using said logical name as
/INCLUDE argument.

A note was added to NOTES.VMS regarding these limitations.

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

Move dso.h to internal

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

Remove several unused undocumented functions.

Removed the following:
    DSO_bind_var, DSO_bind_var, DSO_get_default_method,
    DSO_get_loaded_filename, DSO_get_loaded_filename, DSO_get_method,
    DSO_new_method, DSO_pathbyaddr, DSO_set_default_method, DSO_set_method,
    DSO_set_name_converter, DSO_set_name_converter

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

Fix error code

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

Small fixups in DSO

- VMS configs had no dso_scheme
- Incorrect return of NULL method.

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

Removed unused fields of CT_POLICY_EVAL_CTX

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

Correct another batch of typos

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

Check algo_strength using SSL_STRONG_MASK

algo_strength contains 2 parts that need to be checked by their own.

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
MR: #2082

remove unused references field

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

Have only one DSO_METHOD_openssl

Instead of have every DSO_METHOD_xxx in all platforms, ensure that only
one DSO_METHOD_openssl is available on all platforms.

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

Make DSO opaque.

This was really easy.

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

Make X509_PUBKEY opaque

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

move x_pubkey.c to crypto/x509

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

Fix enable-zlib no-comp.

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

Fix no-rc4.

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

Move declaration of i into blocks where it is used.

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

sctp requires dgram.

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

RT4660: BIO_METHODs should be const.

BIO_new, etc., don't need a non-const BIO_METHOD. This allows all the
built-in method tables to live in .rodata.

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

Downcase VMS config names

On VMS, we downcase option names, which means that config names are
downcased as well, so they need to be downcased in the target table to
be found.

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

Just like bio_out, bio_err needs the linebuffer filter on VMS

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

Adds s2i function for ct_precert_poison X509 extension

Allows CONF files for certificate requests to specify that a pre-
certificate should be created (see RFC6962).

Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Fix memory leak where fdlookup linked list is not freed during
ASYNC_WAIT_CTX_free

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

Fix no-ts with --strict-warnings

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

Fix no-sock

Misc fixes for no-sock

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

Fix no-gost

Configure had the wrong name for the no-gost option.

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

Don't use hardcoded values for types

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

In for loop values, introduce a dummy to protect against empty list

In constructions such as 'for x in $(MAKEVAR); do ...', there's the
possibility that $(MAKEVAR) is en empty value.  Some shells don't like
that, so introduce a dummy value that gets discarded:

    for x in dummy $(MAKEVAR); do
        if [ "$$x" = "dummy" ]; then continue; fi

Closes RT#4459

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

Fix no-seed

Fix compilation with --strict-warnings and no-seed

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

Remove the no-rijndael option

Rijndael is an old name for AES.

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

Fix no-scrypt

Fix the evp tests when no-scrypt is used.

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

Skip the CMS tests if CMS is disabled

This fixes the no-cms compile time option.

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

Fix no-dsa

Misc fixes for no-dsa.

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

Build system: VC-WIN64I fixups.

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

Configurations/windows-makefile.tmpl: respect no-makedepend.

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

Windows build system: get uplink right.

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

config: fix off-by-centimeter-to-the-right typo.

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

Configurations/10-main.conf: freeze -std option in darwin*-ppc-cc.

RT#4422

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

Remove mk1mf documentation

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

Remove the remainder of util/mk1mf.pl and companion scripts

This removes all scripts that deal with MINFO as well, since that's
only used by mk1mf.

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

Remove generation of ms/version32.rc from Configure, use util/mkrc.pl

utils/mkrc.pl was added a while ago as a better generator for the
Windows DLL resource file.  Finalize the change by removing the
ms/version32.rc generator from Configure and adding resource file
support using mkrc.pl in Configurations/windows-makefile.pl

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

Remove the mk1mf VC-WIN* builds and its supporting scripts

The mk1mf build for the VC-WIN* targets is broken and the unified
scheme works well enough, so we clean out the old.

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

Fix ALPN - more fixes

* Clear proposed, along with selected, before looking at ClientHello
* Add test case for above
* Clear NPN seen after selecting ALPN on server
* Minor documentation updates

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

Add a comment on dane_verify() logic

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

Remove #error from include files.

Don't have #error statements in header files, but instead wrap
the contents of that file in #ifndef OPENSSL_NO_xxx
This means it is now always safe to include the header file.

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

GH886: CONNECT should use HTTP/1.1

By default you get 0.9 which isn't widely available.
But we use HTTP/1.0 for now.
Courtesy beusink@users.github.com

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

Have makedepend output to stdout and redirect it

This gives us better control of what files are produced.

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

ct_int.h only needed by crypto/ct

Merge ct_int.h into ct_locl.h

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

constify DSA_SIG_get0()

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

Move blake2_loclh to blake2 directory

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

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

Fixed a bunch of typos in the docs

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

constify ECDSA_SIG_get0()

PR#4436

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

If the asm file to be compiled isn't generated, leave the ext alone

Closes RT#4447

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

make update

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

Don't let 'generate' target depend on generated files, act directly instead

One of the 'generate' targets depended on $(SRCDIR)/apps/progs.h,
which depended on...  nothing.  This meant it never got regenerated
once it existed, regardless of need.  Of course, we could have it
depend on all the files checked to generate it, but they also depend
on progs.h, so we'd end up getting cricular dependencies, which makes
make unhappy.

Furthermore, and this applies for the other generated files, having
them as targets means that they may be regenerated on the fly in some
cases, and since they get written to the source tree, this isn't such
a good idea if that tree is read-only (which is a possible situation
in an out-of-tree build).

So, we move all the actions to the 'generate' targets themselves, thus
making sure they get regenerated in a controlled manner and regardless
of dependencies.

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

apps/progs.pl: add back the INCLUDE_FUNCTION_TABLE wrapper

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

Revert "Ignore the generated apps/progs.h"

This reverts commit 91056e72693b4ee8cb5339d9091871ffc3b6f776.

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

Revert "Generate apps/progs.h on the fly"

This reverts commit 04e2a527379ad12ca512aef4e838f94af22d7f79.

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

Revert "Include progs.h directly in openssl.c instead of via apps.h"

This reverts commit a45d7d5388c6774a484cff4af13f188240d3d50b.

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

GH36: Add casts to match the documentation

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

Correct faulty L<> links in .pods

Closes RT#4450

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

Add a few more quotes for Windows (install_ssldirs)

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

The command source are files, not directories

Therefore, they should be concatenated with the source directory using
catfile(), not catdir()

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

Replace sed command with perl

Some implementations of sed require a newline before an ending '}'.
The easier method is to replace that sed command with the
corresponding perl command.

Closes RT#4448

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

Clear the exit code from 'find' in 'make depend'

Depending on what has been built so far, all .d files may not be
present and 'find' will exit with non-zero exit code.  This isn't a
bother for us but may break make, so clear the exit code with an added
'exit 0'.

Closes RT#4444

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

Shut find up about files not found

Some of the files in $(DEPS) might not exist.  We don't need to know.

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

Don't free up EVP_MD_CTX.

Don't free up passed EVP_MD_CTX in ASN1_item_sign_ctx(). This
simplifies handling and retains compatiblity with previous behaviour.

PR#4446

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Stephen Henson <steve@openssl.org>

Use the dependencies received in generatesrc()

generatesrc() did already receive dependency information, but never
used it, and never really needed to...  until now.

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

Ignore the generated apps/progs.h

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

Generate apps/progs.h on the fly

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

Include progs.h directly in openssl.c instead of via apps.h

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

Make apps/progs.pl more flexible

Make Configure write @disablables to configdata.pm and have
apps/progs.pl use that data.

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

Correct the export list when writing configdata.pm

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

Correct the check of files to depend on against known generated ones

The check was erroneously inversed

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

Don't perform tsa tests if configured "no-ts"

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

Correct incorrect path

In crypto, buildinf.h depends on ../configdata.pm, not ./configdata.pm

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