Re: [Libevent-users] How to set "allow_dirty_shutdown" for SSL?

[Joachim Bauch <jojo@xxxxxxxxxxx>]

Hi,

Nick Mathewson wrote:
[...]
> At this point, I don't recall whether I started adding that feature
> and didn't finish, or started removing it and didn't finish.  :/

I see :)

> The SSL clean-shutdown protocol exists so that you can tell the
> difference between the other side closing the connection correctly,
> and an attacker forcing the connection to close.  For some protocols
> (like ones that use a close to indicate the end of a data object) this
> can matter, since otherwise you have no way to tell whether the other
> side meant to stop sending you data, or whether it got cut off.  If
> you've got one of those protocols, treating a broken connection the
> same way as a closed connection is a big security hole.  You might as
> well downgrade to SSL2. ;)
>
> On the other hand, there should be _some_ way of detecting and
> reporting a dirty shutdown more detectable.  Instead of having an
> allow_dirty_shutdown flag, maybe there should be a got_dirty_shutdown
> flag that the user can check when getting an SSL error.  (There
> doesn't seem to be a particular OpenSSL error that gets set in this
> case, so I don't expect that calling bufferevent_get_openssl_error
> would do any good.)

Okay, understood. My problem is, that for some clients the TCP
connection is closed without the SSL-close protocol being performed
beforehand. This is not really an error in my use case, but I can't
differentiate between this dirty shutdown and a regular error - both
events are reported as "BEV_EVENT_ERROR" to my app.

My application-level callback is not triggered from "conn_closed()"
as Jardel wrote, but from "be_openssl_eventcb" (my SSL bufferevent is
wrapped around a regular bufferevent).

To report a dirty shutdown, I suggest passing an event of
"BEV_EVENT_ERROR|BEV_EVENT_EOF" to the application (instead of the
"BEV_EVENT_ERROR" as is now), so the dirty shutdown can be detected
and handled differentely to the clean shutdown (which is "BEV_EVENT_EOF"
only).

Any objections or should I prepare a patch for this?

Best regards,
  Joachim

***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxx with
unsubscribe libevent-users    in the body.