Re: Who's at fault here?

[Nicolas Williams <Nicolas.Williams%sun.com@localhost>]

On Wed, Nov 30, 2005 at 09:30:51PM -0700, Joseph Galbraith wrote:
> Nicolas Williams wrote:
> > On Fri, Oct 21, 2005 at 12:26:29AM -0400, der Mouse wrote:
> >> The thing is, the implementation I'm talking to (which is bannering as
> >> "SSH-1.99-OpenSSH_3.4 NetBSD_Secure_Shell-20020626") is logging
> >> "Received window adjust for non-open channel 0." in response to this.
> > 
> > This sort of thing can happen with "window-change" channel requests
> > received before the shell/exec/subsystem channel requests on the same
> > channel.
> 
> But it shouldn't, should it?

Indeed, it shouldn't, by my reading.

> My reading of the draft has always been it is legal to send a
> window adjust any time the channel is open.
> 
> Wait... did you say "window-change" ?  As in the size of the
> screen changed message.  Are you saying the error is ...

I meant that this type of error happens when "window-change" channel
requests...  Not that that very same error message message is elicited.

> unclear ... and should really read "Received window-change
> for non-open channel 0?" and it occurs because a "window-change"
> was sent before pty has been allocated?

The issue is that the window-change request may be sent before the
shell/exec/subsystem request but after the pty-req request.

Our bug ID for it is:

6336266 ssh SIGWINCH handling races; some servers are very unforgiving

Nico
--