13 May 2012 oubiwann   » (Journeyer)

Twisted SSH: Rendering a Log-in Banner/MOTD in Conch

A few weeks ago, I pinged my peeps on #twisted asking why the banner for a custom SSH server wasn't rendering properly. After some digging around and some inconsistent results (well, consistently bad results for me), we weren't able to resolve anything, and I had to set the problem aside.

The Symptom
The first thing I had tried was subclassing Manhole from twisted.conch.manhole, overriding (and up-calling) connectionMade, writing the banner to the terminal upon successful connection. This didn't work, so I then tried overriding initializeScreen by subclassing twisted.conch.recvline.RecvLine. Also a no-go. And by "didn't work" here's what I mean:

In both Linux (Ubuntu 12.04 LTS, gnome-terminal) and Mac (OS X 10.6.8, Terminal.app), after a successful login to the Twisted SSH server, the following sequence would occur:
  1. an interactive Python prompt was rendered, e.g., ":>>"
  2. the banner was getting written to the terminal, and
  3. the terminal screen refreshed with the prompt at the top
This all happened so quickly, that I usually never even saw #1 and #2. Just the second ":>>" prompt from #3. Only by scrolling up the terminal buffer would I see that the banner had actually been rendered. Even though I was doing my terminal.write after connectionMade and initializeScreen, it didn't seem to matter.

Some time last week, I put together example Twisted plugins showing what the problem was, and the circumstances under which a banner simply didn't get rendered. The idea was that I would provide some bare-bones test cases that demonstrated where the problem was occurring, post them to IRC or the Twisted mail list, and we could finally get it resolved. 'Cause, ya know, I really want my banners ...

While tweaking the second Twisted plugin example, I finally poked my head into the right method and discovered the issue. Here's what's happening:

  • twisted.conch.recvline.RecvLine.connectionMade calls t.c.recvline.RecvLine.initializeScreen
  • t.c.recvline.RecvLine.initializeScreen does a terminal.reset, writes the prompt, and then switches to insert mode. But this is a red herring. Since something after initializeScreen is causing the problem, we really need to be asking "who's calling connectionMade?"
  • t.c.manhole_ssh.TerminalSession.openShell is what kicks it off when it calls the transportFactory (which is really TerminalSessionTransport)
  • openShell takes one parameter, proto -- this is very important :-)
  • openShell instantiates TerminalSessionTransport
  • TerminalSessionTransport does one more thing after calling the makeConnection method on an insults.ServerProtocol instance (the one I had tried overriding without success), and as such, this is the prime suspect for what was preventing the banner from being properly displayed: it calls  chainedProtocol.terminalProtocol.terminalSize
  • chainedProtocol is an insults.ServerProtocol instance, and its terminalProtocol attribute is set when ServerProtocol.connectionMade is called.
  • A quick check reveals that terminalProtocol is none other than the proto parameter passed to openShell.

But what is proto? Some debugging (and the fact that of the three terminalSize methods in all of twisted, only one is an actual implementation) reveals that proto is a RecvLine instance. Reading that method uncovers the culprit in our whodunnit:  the first thing the method does is call terminal.eraseDisplay.

Bingo! (And this is what I was referring to above when I said "poked my head" ...)

Since this was called after all of my attempts to display a banner using both connectionMade and initializeScreen, there's no way my efforts would have succeeded.

Here's What You Do
How do you get around this? Easy! Subclass :-)

The class  TerminalSessionTransport in t.c.manhole_ssh is the bad boy that calls terminalSize (which calls eraseDisplay). It's the last thing that TerminalSessionTransport does in its __init__, so if we subclass it, and render our banner at the end of our __init__, we should be golden. And we are :-)

You can see an example of this here.

Not sure if this sort of thing is better off in projects that make use of Twisted, or if it would be worth while to add this feature to Twisted itself. Time (and blog comments) will tell.

As is evident from the screenshot above (and the link), this feature is part of the DreamSSH project. There are a handful of other nifty features/shortcuts that I have implemented in DreamSSH (plus some cool ones that are coming) and I'm using them in projects that need a custom SSH server. I released the first version of DreamSSH last night, and there's a pretty clear README on the github project page.

One of the niftier things I did last night in preparation for the release was to dig into Twisted plugins and override some behaviour there. In order to make sure that the conveniences I had provided for devs with the Makefile were available for anyone who had DreamSSH installed, I added subcommands... but if the service was already running, these would fail. How to work around that (and other Twisted plugin tidbits) are probably best saved for another post, though :-)

Syndicated 2012-05-13 00:50:00 (Updated 2013-04-15 04:25:41) from Duncan McGreggor

Latest blog entries     Older blog entries

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!