Handle TLS-intolerant servers by retrying with TLS 1.0...

Handle TLS-intolerant servers by retrying with TLS 1.0
turned off.

R=darin
BUG=3001
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=3017

[darin (slow to review), 2008-09-30 20:59:31]

LGTM

http://codereview.chromium.org/5617/diff/11/214
File net/base/client_socket_factory.h (right):

http://codereview.chromium.org/5617/diff/11/214#newcode27
Line 27: unsigned int protocol_versions) = 0;
nit: this third parameter is not very obvious given its type or name, so we
probably should have some comments explaining what it is.

http://codereview.chromium.org/5617/diff/11/218
File net/base/ssl_client_socket.h (right):

http://codereview.chromium.org/5617/diff/11/218#newcode46
Line 46: unsigned int protocol_versions);
nit: just make this an int valued parameter and name it protocol_version_mask? 
"int" is usually a good choice unless there is strong need for unsigned.

http://codereview.chromium.org/5617/diff/11/210
File net/http/http_network_transaction.cc (right):

http://codereview.chromium.org/5617/diff/11/210#newcode63
Line 63: ssl_versions_ = SSLClientSocket::SSL3 | SSLClientSocket::TLS1;
hmm... doesn't have to happen in this CL, but i think we should find a way to
have these values defined on all platforms.

http://codereview.chromium.org/5617/diff/11/211
File net/http/http_network_transaction.h (right):

http://codereview.chromium.org/5617/diff/11/211#newcode190
Line 190: unsigned int ssl_versions_;
we can't use this member on all platforms?

[wtc, 2008-10-01 01:01:59]

Please review the new snapshot.

Here are my responses to your comments.

protocol_versions has been renamed protocol_version_mask,
and ssl_versions has been renamed ssl_version_mask.

I added a comment to client_socket_factory.h to explain the
protocol_version_mask parameter.

I used unsigned int for the SSL versions bitmask after
consulting the Google C++ Style Guide and checking with
brettw and mmentovai.  But I have changed it to int to
follow the local convention in the net module.  (In
fact I used int originally.)

The Mac or Linux porting team added #if defined(OS_WIN)
around #include "net/base/ssl_client_socket.h", which is
why I had to slap #if defined(OS_WIN) around anything
that uses the SSL versions bitflags defined in that header.

[darin (slow to review), 2008-10-01 01:08:22]

LGTM

http://codereview.chromium.org/5617/diff/38/245
File net/base/ssl_client_socket_unittest.cc (right):

http://codereview.chromium.org/5617/diff/38/245#newcode17
Line 17: const unsigned int kSSL3TLS1 = net::SSLClientSocket::SSL3 |
maybe it would be good to call this kUseSSL3OrTLS1?  or maybe
kDefaultSSLVersionMask?

[wtc, 2008-10-01 01:13:17]

I like kDefaultSSLVersionMask.  I will change kSSL3TLS1 to
kDefaultSSLVersionMask before checking in the CL.