This patch fixes a bug in Happy Eyeballs where curl would wait for a
connect response from socket1 before checking socket2.

Also, it updates error messages for failed connections, showing the ip
addresses that failed rather than just the host name repeatedly.

Bug: http://curl.haxx.se/mail/lib-2013-10/0236.html
Reported-by: Paul Marks

This patch invokes two socket connect()s nearly simultaneously, and
the socket that is first connected "wins" and is subsequently used for
the connection. The other is terminated.

There is a very slight IPv4 preference, in that if both sockets connect
simultaneously IPv4 is checked first and thus will win.

Should a client application fail to decode an authentication message
received from a server, or not support any of the parameters given by
the server in the message, then the authentication phrase should be
cancelled gracefully by the client rather than simply terminating the
connection.

The authentication phrase should be cancelled by simply sending a '*'
to the server, in response to erroneous data being received, as per
RFC-3501, RFC-4954 and RFC-5034.

This patch adds the necessary state machine constants and appropriate
response handlers in order to add this functionality for the CRAM-MD5,
DIGEST-MD5 and NTLM authentication mechanisms.

...in preparation for upcoming modifications.

warning: 'result' may be used uninitialized in this function

This is a regression since the switch to always-multi internally
c4312741.

Test 1316 was modified since we now clearly call the Curl_client_write()
function when doing the LIST transfer part and then the
handler->protocol says FTP and ftpc.transfertype is 'A' which implies
text converting even though that the response is initially a HTTP
CONNECT response in this case.

This workaround had been previously been implemented for IMAP and POP3
but not SMTP. Some of the recent test case additions implemented this
behaviour to emulate a bad server and the SMTP code didn't cope with it.

Corrected 80 character line length error and pointer declarations (some
of which were previously incorrect)

IFS compilation support, SSL GSKit backend by default, TLSv1.[12] support in
  GSKit for OS400 >= V7R1, no more tabs in make scripts.

error: unused variable 'table16'

... if not already initialized.  This fixes a regression introduced by
commit 4ad8e142, which caused test619
to intermittently fail on certain machines (namely Fedora build hosts).

Changed the failure code when TLS v1.1 and v1.2 is requested but not
supported by older OpenSSL versions, following review from libcurl
peers, and reduced the number of required preprocessor if statements.

...with the use of CURL_SSLVERSION_TLSv1_1 and CURL_SSLVERSION_TLSv1_2
being conditional on OpenSSL v1.0.1 as the appropriate flags are not
supported under earlier versions.

Commit ad34a2d5 relies on definitions that are only present in
OpenSSL v1.0.1 and up. This quick fix allows the builds that use
older versions of OpenSSL to continue building.

According to the documentation for libssh2_userauth_list(), a NULL
return value is not necessarily an error. You must call
libssh2_userauth_authenticated() to determine if the SSH_USERAUTH_NONE
request was successful.

This fixes a segv when using sftp on a server that allows logins with an
empty password. When NULL was interpreted as an error, it would
free the session but not flag an error since the libssh2 errno would be
clear. This resulted in dereferencing a NULL session pointer.

Signed-off-by: Tyler Hall <tylerwhall@gmail.com>

CURL_SSLVERSION_TLSv1_0, CURL_SSLVERSION_TLSv1_1,
CURL_SSLVERSION_TLSv1_2 enum values are added to force exact TLS version
(CURL_SSLVERSION_TLSv1 means TLS 1.x).

axTLS:
axTLS only supports TLS 1.0 and 1.1 but it cannot be set that only one
of these should be used, so we don't allow the new enum values.

darwinssl:
Added support for the new enum values.

SChannel:
Added support for the new enum values.

CyaSSL:
Added support for the new enum values.
Bug: The original CURL_SSLVERSION_TLSv1 value enables only TLS 1.0 (it
did the same before this commit), because CyaSSL cannot be configured to
use TLS 1.0-1.2.

GSKit:
GSKit doesn't seem to support TLS 1.1 and TLS 1.2, so we do not allow
those values.
Bugfix: There was a typo that caused wrong SSL versions to be passed to
GSKit.

NSS:
TLS minor version cannot be set, so we don't allow the new enum values.

QsoSSL:
TLS minor version cannot be set, so we don't allow the new enum values.

OpenSSL:
Added support for the new enum values.
Bugfix: The original CURL_SSLVERSION_TLSv1 value enabled only TLS 1.0,
now it enables 1.0-1.2.

Command-line tool:
Added command line options for the new values.

SecPKCS12Import() returns a few errors that are enumerated in OS X's
headers but not in iOS' headers for some reason.

Reported-by: Petr Pisar

Write to CURLOPT_ERRORBUFFER information about mismatch alternative
certificate subject names.

Signed-off-by: Andrej E Baranov <admin@andrej-andb.ru>

This fixes a regression introduced by commit 0feeab78 limiting the speed
of SCP upload to 16384 B/s on a fast connection (such as localhost).

lib/curl_setup_once.h assumed lwIP on Windows uses 'SetLastError()' to
set network errors. It doesn't; it uses 'errno'.

When using the -w '%{http_code}' flag and simulating a Not Modified then
304 should be output.

Credit (for catching a cipher I forgot to add to the blocked ciphers list):
https://www.ssllabs.com/ssltest/viewMyClient.html

Setting only CURLOPT_SSL_VERIFYHOST without CURLOPT_SSL_VERIFYPEER set
should still verify that the host name fields in the server certificate
is fine or return failure.

Bug: http://curl.haxx.se/mail/lib-2013-10/0002.html
Reported-by: Ishan SinghLevett