[tbnl-devel] Re: TBNL file descriptor leak

[Edi Weitz]

On Sun, 22 Jan 2006 10:20:04 -0700, "Robert J. Macomber" <tbnl at rojoma.com> wrote:

> In my apache2 config, I've got this [simplified]:
>
> LispServer 127.0.0.1 3000 lisp
> <VirtualHost *:80 *:54321>
> 	ServerName lisp.rojoma.com
>         DocumentRoot /home/robertm/mod_lisp
>         <LocationMatch /.*\.l>
>                 SetHandler lisp-handler
>         </LocationMatch>
> </VirtualHost>
>
> If you visit the url http://lisp.rojoma.com/%ef.l apache sends bytes
> for the script-filename parameter which can be interpreted as the
> latin-1 string "/home/robertm/mod_lisp/ï.l".  If they're interpreted
> as UTF-8, like sbcl's READ-LINE tries to do when running in an UTF-8
> locale (since kmrcl doesn't specify an explicit :external-format in
> ACCEPT-TCP-CONNECTION) an error is thrown in GET-REQUEST-DATA.
>
> In fact, a user can make apache send arbitrary bytes to lisp while
> it's reading the request headers in other ways too (e.g., sending a
> "Mumble: {random octets}" header) but script-filename is the only
> one where Apache will do so more or less on its own.
>
> I changed kmrcl here rather than making tbnl send a "bad request"
> reply because you never know what random junk proxies and such will
> throw into users' HTTP request headers, and I didn't want to deny
> service just because some proxy is adding the (accented) name of
> their ISP to user-agent strings or something.  I haven't actually
> seen that happen yet, but the net is vast.

OK, thanks for the info.  I've been thinking about extracting the
relevant bits from KMRCL and integrating them into TBNL directly
anyway.  We only use a small fraction of what's in there and it might
make sense to make some of this user-configurable, not only because of
the external format but also because of the interface(s) TBNL will
listen on, for example.  I'll do that one day... :)

Cheers,
Edi.