[Mono-dev] Re: [PATCH] Fully Asynchronous and Re-Factored Ssl Streams in Mono.Security

[Sebastien Pouliot]

Hello JD,

On Wed, 2005-24-08 at 21:12 -0700, JD Conley wrote:
> I took the plunge and fully implemented async Ssl streams for both
> client and server.  This fixes
> http://bugzilla.ximian.com/show_bug.cgi?id=75687 as well as other bugs
> I've been talking with Sebastien and Carlos about off list. This patch
> passes the SocketHell tests that I contributed to them last week
> (multi threaded concurrent async streaming through TCP sockets).

Wow! That's the kind of surprise I like when I wake up :-)

> It's also a bit of a re-factor, though I only touched two classes and
> added one.  I pulled most of the code out of the individual
> SslClientStream and SslServerStream and moved it down into a new
> abstract SslStreamBase.  Client and server are now practically the
> same implementation.  I left all existing interfaces present, but
> obviously had to add some new ones.

We always "sticked" to the Fx 1.2 preview specs for the Ssl[Client|
Server]Stream API and didn't planned to change this before implementing
the new SslStream (in System.dll) for 2.0 - BUT this isn't a major
problem if it doesn't break binary compatibility (for current
applications linked with Mono.Security.dll).

> The only "gotcha" is if you start running low on threadpool threads.
> Then the Ssl Stream will fall back to a synchronous handshake. The
> other option here is to spawn a thread ourselves for the handshake
> instead of using Delegate.BeginInvoke(), use the IO ThreadPool (is
> that available to Mono.Security?), or just live with a synchronous
> handshake. Of course, 99.999% of the time this issue won't occur and
> won't be a problem unless you have both client and server sharing the
> same Threadpool causing a deadlock.

That's not worse than the original (where the handshake was always
synchronous).

> I hope this helps and gets integrated.  It's definitely necessary for
> our implementation.

1. I'll review the patch and, in doing so, check for possible binary
compatibility problems. I'm sure Carlos will do likewise;

2. I'll make public a new Mono.Security.dll assembly so that people that
depends on Ssl*Stream may tests it and report any problem/difference;

3. I'll run the regressions tests, the tools under /mcs/class/
Mono.Security/Test/tools/*, to see if it works in previously reported
conditions.

4. Commit (both the patch and your SocketHell tests). Hopefully we can
do all this before the next release.

Many thanks!
-- 
Sebastien