Submitted by: steve@openssl.org

The original fix for PR#2125 broke compilation on some Unixware platforms:
revert and make conditional on VMS.

length parameter.

ctrl is incorrectly implemented (e.g. some versions of Apache). As a workaround
call both BIO_CTRL_INFO and BIO_CTRL_WPENDING if it returns zero. This should
both address the original bug and retain compatibility with the old behaviour.

server attempts to renegotiate with an unpatched client.

algorithms not found when an application uses PKCS#12 and only calls
SSL_library_init() instead of OpenSSL_add_all_algorithms(). Simple
work around is to add the missing algorithm (40 bit RC2) in
SSL_library_init().

Submitted by: "NARUSE, Yui" <naruse@airemix.jp>

Remove non-ASCII comment which causes compilation errors on some versions
of VC++.

Submitted by: Robin Seggelmann <seggelmann@fh-muenster.de>

Better fix for PR#2144

needed).

PR: 2094,2095

in performance calculations. For the moment there is only one code
fast enough to suffer from this: Intel AES-NI engine.
PR: 2096

crypto/bn/asm/mips3.s from IRIX. Performance improvement is not as
impressive as with complete assembler, but still... it's almost 2.5x
[on R5000].

less).

Submitted by: Mike Frysinger <vapier@gentoo.org>

Change missed references to lib to $(LIBDIR)

Submitted by: steve@openssl.org

Fix DTLS connection so new_session is reset if we read second client hello:
new_session is used to detect renegotiation.

Submitted by: steve@openssl.org

Add missing DTLS state strings.

Submitted by: "Alon Bar-Lev" <alon.barlev@gmail.com>

Fix gcc-aix compilation issue.

Although it will be many years before TLS v2.0 or later appears old versions
of servers have a habit of hanging around for a considerable time so best
if we handle this properly now.

stops applications that call CRYPTO_free_all_ex_data() prematurely leaking
memory.

Submitted by: Willy Weisz <weisz@vcpc.univie.ac.at>

Add options to output hash using older algorithm compatible with OpenSSL
versions before 1.0.0

1. Send SCSV is not renegotiating, never empty RI.
2. Send RI if renegotiating.

HP-UX thing).

connect() are as finicky as sendto() when it comes to socket address
length. As it turned out they are, therefore the fix. Note that you
can't reproduce the problem on Linux, it was failing on Solaris,
FreeBSD, most likely on more...

PR: 2114
Submitted by: Robin Seggelmann

not to mention that first sockaddr_storage had __ss_family, not ss_family.
In other words it makes more sense to avoid sockaddr_storage...

1. Add provisional SCSV value.
2. Don't send SCSV and RI at same time.
3. Fatal error is SCSV received when renegotiating.