[
https://issues.apache.org/jira/browse/QPID-3514?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13121323#comment-13121323
]
[email protected] commented on QPID-3514:
-----------------------------------------------------
bq. On 2011-10-05 15:39:54, Andrew Stitcher wrote:
bq. > ** We can't accept this as it stands as it will break IPv6 support if
turned on. **
bq. >
bq. > On the whole I think it is moving in the correct direction, but it is
only partly formed:
bq. >
bq. > * This works in effect by adding in *another* transport type - the muxed
SSL transport. We are trying to reduce and simplify the number of transports
not increase them, this will make the code harder to maintain over all.
bq. >
bq. > * It's not clear to me why we wouldn't want this always turned on if we
have the capability. In other words what is the benefit of ever turning it off.
When we can do this switch then we should have it on for all listening sockets,
the only option should be to allow a port to only accept encrypted connections.
bq. >
bq. > * This code is tactical, not strategic in that it works for this
particular case, but as far as I can tell won't help in the upcoming amqp 1.0
case where we would need to switch to TLS processing sometime during the amqp
protocol processing.
bq. >
bq. > * The code that selects between encrypted and non encrypted code paths
should be unified with the code that checks the protocol version, currently the
accept code does this and blocks whilst reading some bytes from the connection.
I really don't like this approach.
bq. >
bq. > - Actually this code makes me wonder if we should change the
architecture of the socket code to have an explicit piece of code which is run
just after accepting a socket to select which bits of code get run next.
bq. >
bq.
bq. Gordon Sim wrote:
bq. Is IPv6 supported for SSL at present? If not, and if you view this as
adding the (optional) ability to serve plain sockets on the SSL enabled port,
then one could argue its not really a regression. We would need to add IPv6
support to SSL anyway and it would seem to be to be largely independent of this
change. If however we do support IPv6 on SSL and this breaks it then I would
certainly agree with you.
bq.
bq. While I think its fair to characterise this as tactical, I don't think
it makes a strategic solution any harder. Though it adds an additional protocol
factory it re-uses code very well (better indeed than what is there at
present). It would certainly need modifications for AMQP 1.0, but compared to
the rest of the work required there I don't think that is particularly
significant, nor do I think it makes things any worse.
bq.
bq. The strategic changes around this part of the code are more
fundamental. Until there is something concrete planned there I think it would
be a real shame to bar adding useful new features, providing they don't
aggravate the situation (and I would argue this patch does not).
bq.
bq. Your point around the blocking in the accept is a good one. It would
be much nicer to avoid blocking if possible. Given the dearth of documentation
in this area, would you be willing to flesh out your preferred approach in a
bit more detail?
bq.
bq.
bq. Andrew Stitcher wrote:
bq. I disagree in principle with the idea that adding more code when we
really need to simplify the code doesn't make our job going forward any harder.
bq.
bq. The code in this area is just within my ability to comprehend, adding
more just makes it harder for me. So while this is a reasonably neat tactical
job I strongly suggest that including this code before simplifying what already
there will make it harder to simplify, I know it will make it harder for me to
simplify.
As a note my plan for making SSL run over IPv6 does indeed feature some of the
simplification I'm talking about, specifically I'm introducing a Socket
interface implemented by both SSL and regular TCP sockets so that the higher
level code can be reduced to a single implementation.
This approach also allows for handing over a socket to the SSL code from the
TCP code during the negotiation if required.
So overall my approach to this would be opposite make the TCP code detect SSL
connections and create the SSL wiring, rather than have a specific option to
make the SSL code allow for regular TCP connections.
- Andrew
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/2206/#review2342
-----------------------------------------------------------
On 2011-10-05 13:36:58, Gordon Sim wrote:
bq.
bq. -----------------------------------------------------------
bq. This is an automatically generated e-mail. To reply, visit:
bq. https://reviews.apache.org/r/2206/
bq. -----------------------------------------------------------
bq.
bq. (Updated 2011-10-05 13:36:58)
bq.
bq.
bq. Review request for Andrew Stitcher, michael goulish, Ted Ross, and Chug
Rolke.
bq.
bq.
bq. Summary
bq. -------
bq.
bq. Allow SSL and non-SSL connections on the same port. Patch contributed by
Zane Bitter for use by the Matahari Project.
bq.
bq.
bq. This addresses bug QPID-3514.
bq. https://issues.apache.org/jira/browse/QPID-3514
bq.
bq.
bq. Diffs
bq. -----
bq.
bq. /trunk/qpid/cpp/src/qpid/sys/ssl/SslSocket.cpp 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/ssl/SslIo.cpp 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/ssl/SslSocket.h 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/ssl/SslIo.h 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/TCPIOPlugin.cpp 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/Socket.h 1179157
bq. /trunk/qpid/cpp/src/qpid/sys/SslPlugin.cpp 1179157
bq.
bq. Diff: https://reviews.apache.org/r/2206/diff
bq.
bq.
bq. Testing
bq. -------
bq.
bq. make check passes; tested with and without the multiplexing; tested client
authentication; tested SASL EXTERNAL (requires reverting a recent change
unrelated to this, see QPID-3522)
bq.
bq.
bq. Thanks,
bq.
bq. Gordon
bq.
bq.
> Allow SSL and non-SSL connections on the same port
> --------------------------------------------------
>
> Key: QPID-3514
> URL: https://issues.apache.org/jira/browse/QPID-3514
> Project: Qpid
> Issue Type: Improvement
> Components: C++ Broker
> Affects Versions: 0.12
> Reporter: Zane Bitter
> Attachments: ssl-mux.patch
>
>
> The Matahari Project (http:://matahariproject.org for the uninitiated) has
> run into an issue with our use of Qpid in that IANA policy is now to refuse
> to assign separate TCP ports for SSL/TLS-wrapped versions of protocols, which
> leaves us with only a single port assigned to Matahari.
> We would like to be able to accept both SSL and non-SSL connections on the
> same port.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
