[Xcb] Re: new SELinux protocol reply problem

[Jeremy A. Kolb]

On Sat, 24 Feb 2007, Barton C Massey wrote:

> In message <20070225001704.GA8949 at id.minilop.net> you wrote:
> > On Sat, Feb 24, 2007 at 11:06:00AM -0800, Barton C Massey wrote:
> > > Is there any reason XCB shouldn't accept short replies?  As
> > > near as I can tell, violating the protocol spec in this
> > > particular way shouldn't ought to hurt anything; JG
> > > apparently agrees?
> > 
> > X11 does not allow encoding replies shorter than 32 bytes without
> > padding, because the length of a reply in bytes is defined as
> > 	32 + reply.length * 4
> 
> Right. We'd probably have to decide that the length is
> signed, which would break a bunch of existing protocol, or
> we'd have to do something equally clever.  It's too late for
> that---my bad.
> 
> > I think JG's comment was somewhat confusing as it was only completely
> > accurate for events and errors (exactly 32 bytes), not replies (at least
> > 32 bytes). The historical reasons behind these protocol details are
> > always interesting though.
> 
> I think the small-size problems were probably also the
> reason for the wacky reply limitation.
> 
> > > Maybe we could reject this protocol unless some explicit
> > > "short-reply" attribute were set in the XML?  I doubt this
> > > will be the last time the problem occurs...
> > 
> > But nothing was wrong on the client side, so why would we reject it? The
> > server sent a length that didn't match the number of bytes it
> > transmitted, which is simply a server bug.
> 
> I guess the XML compilation warning would be if you defined
> a reply type that was less than 32 bytes without either
> adding padding or the special "short-reply" attribute; marginal
> plan at best, but might keep that mistake from happening
> again.  On the other hand, it would noise the XML some for
> marginal gain.  Overall, I'll withdraw the suggestion.
>

I was under the assumption that we didn't need to specify padding at the 
end of a reply because it was taken care of.  I think the server-side xcb 
would do the same.

> 
> Thanks much for the clarifications.
> 
>     Bart
> _______________________________________________
> Xcb mailing list
> Xcb at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/xcb
>