Implement SSL renegotiation....

Implement SSL renegotiation.

In the Windows Schannel API, a server requests renegotiation
when DecryptMessage (decrypting received data) returns
SEC_I_RENEGOTIATE.  We need to jump to the handshake
sequence, and when handshake completes, come back to reading
data.

I also cleaned up the code.  I created the
SetNextStateForRead and FreeSendBuffer functions to share
common code, and made sure our handshake sequence is
completely equivalent to the handshake sequence in the
Platform SDK WebClient.c sample.

R=rvargas
BUG=6893
TEST=Visit these sites, which request SSL renegotiation:
https://secure.skandiabanken.se/Skbsecure/LoginInternet/SKBLoginInternet.aspx
https://secure.skandiabanken.no/SkbSecure/Authentication/Otp/Default.ashx
https://www.myopenid.com/signin_certificate

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=12229

[wtc, 2009-03-20 21:24:59]

http://codereview.chromium.org/42380/diff/2002/2004
File net/base/ssl_client_socket_win.cc (left):

http://codereview.chromium.org/42380/diff/2002/2004#oldcode562
Line 562: in_buffers[0].pvBuffer = &recv_buffer_[0];
The change from &recv_buffer_[0] to recv_buffer_.get() is
a no-op.  It is for consistency with the rest of the file.

http://codereview.chromium.org/42380/diff/2002/2004#oldcode592
Line 592: if (status == SEC_E_INCOMPLETE_MESSAGE) {
Starting here, I did a thorough comparison with the PSDK
WebClient.c sample to make sure our code is completely
equivalent.

[rvargas (doing something else), 2009-03-20 23:07:16]

Quite interesting. LGTM.

http://codereview.chromium.org/42380/diff/2002/2004
File net/base/ssl_client_socket_win.cc (right):

http://codereview.chromium.org/42380/diff/2002/2004#newcode592
Line 592: FAILED(isc_status_) && (out_flags & ISC_RET_EXTENDED_ERROR))) {
nit: Do you mind having an extra () around this line?. It makes me a little
nervous to have a series of || and && without parenthesis.

http://codereview.chromium.org/42380/diff/2002/2004#newcode827
Line 827: return ERR_SSL_RENEGOTIATION_REQUESTED;
nit: Maybe we could have a different error code here so that we know if we have
to implement this (assuming that we receive bug reports with just the error code
and not a log).

http://codereview.chromium.org/42380/diff/2002/2003
File net/base/ssl_client_socket_win.h (right):

http://codereview.chromium.org/42380/diff/2002/2003#newcode133
Line 133: // false if we're wrting a subsequent token.  After we have written a
token
nit: writing

[wtc, 2009-03-20 23:34:15]

Ricardo,

Thanks for the review.  I made the changes you suggested
except adding new error codes for the conditions that I
assume can't happen.  I've been very successful with
asking users to run chrome.exe with --enable-logging
and pasting the chrome_debug.log contents, so the LOG(ERROR)
statements should be adequate.

I've checked in this CL.