lingering close on aborted connections

["Roy T. Fielding" <fi...@kiwi.ics.uci.edu>]

Right now we are not bothering to do a lingering close on an
aborted connection.  The reasoning was that this type of error
might be due to a non-HTTP request or a deliberate attempt to
crack the server via a buffer overload, and therefore it wasn't
worth the effort of ensuring the client got the response.

However, this leads to testing and behavioral problems if the
client is not attempting to hack the server.  Like the PUT issue,
the client is sending too much message/URI data, and thus
won't see the HTTP error response if it receives a TCP RST before
it can get an ACK packet for that response back to the server.
Thus, the client will just try the same broken request again, ...

Should I fix this in http_main.c so that we will do a lingering
close in these cases (basically, on any connection in which we have
sent an HTTP response for the current request)?

....Roy

Re: lingering close on aborted connections

[Rodent of Unusual Size <Ke...@Golux.Com>]

Roy T. Fielding wrote:
> 
> Should I fix this in http_main.c so that we will do a lingering
> close in these cases (basically, on any connection in which we have
> sent an HTTP response for the current request)?

I say 'aye.'

#ken	P-)}

Ken Coar                    <http://Web.Golux.Com/coar/>
Apache Group member         <http://www.apache.org/>
"Apache Server for Dummies" <http://Web.Golux.Com/coar/ASFD/>