OpenSSL: add support for the TLS padding extension.

OpenSSL: add support for the TLS padding extension.

This worksaround a bug in some versions of F5 devices that causes the
connection to hang when the ClientHello record is between 256 and 511 bytes
long.

https://tools.ietf.org/html/draft-agl-tls-padding-02

Based on the upstream patch to do the same thing:
0467ea68624450ecece4cde0d5803499aaff19c2

BUG=none

[agl, 2013-12-12 16:46:09]

Note: have changed the CL to fix a bug. There are two places where OpenSSL
creates a ClientHello: in s23_clnt.c and s3_clnt.c. The former includes the
record header in the buffer but that latter does not. The code now handles both
cases correctly.

[digit1, 2013-12-12 17:02:21]

ltgm

[digit1, 2013-12-12 17:02:28]

https://codereview.chromium.org/112933006/diff/20001/openssl/ssl/t1_lib.c
File openssl/ssl/t1_lib.c (right):

https://codereview.chromium.org/112933006/diff/20001/openssl/ssl/t1_lib.c#new...
openssl/ssl/t1_lib.c:670: * not. */
That's really subtle. Is this per-spec/per-protocol or an implementation detail?
If the latter, how about adding a comment in the appropriate places to avoid
breaking this if the sources are independently modified in the future?

[agl, 2013-12-16 19:44:19]

https://codereview.chromium.org/112933006/diff/20001/openssl/ssl/t1_lib.c
File openssl/ssl/t1_lib.c (right):

https://codereview.chromium.org/112933006/diff/20001/openssl/ssl/t1_lib.c#new...
openssl/ssl/t1_lib.c:670: * not. */
On 2013/12/12 17:02:28, digit1 wrote:
> That's really subtle. Is this per-spec/per-protocol or an implementation
detail?
> If the latter, how about adding a comment in the appropriate places to avoid
> breaking this if the sources are independently modified in the future?

This is sadly an implementation niggle. I've added comments at the sites where
the buffers are setup about the subtle dependency.