[ECOS] FreeBSD socket loss (a.k.a. MSIE DoS attack)
Gary Thomas
gary@mlbassoc.com
Tue Apr 17 11:02:00 GMT 2007
Lars Povlsen wrote:
> Hi All!
>
> I have run into a weird occurrence of total socket buffer drainage with
> the FreeBSD network stack (IPv4).
>
> The problem is triggered by MSIE going bezerk while rendering an
> Ajax/DOM style, graphics heavy, web page. It goes into a mode of a
> series of spastic connnects to the (eCos) HTTP server, request a graphic
> object, followed immediately by a RST. Then a new connection, etc. The
> browser manages to get the job done, at the expense of TCP connections
> and - worse - the FreeBSD stack loosing socket buffers - forever!
>
> When the network stack is void of buffers, exiting the browser only
> frees *1* socket buffer. And waiting > 10 minutes does not uncover more
> buffers from the depths of the stack.
>
> While I realize that MSIE is acting up - BIG TIME - I hate that it can
> cause semi-permanent damage to the operation of my system. Does anybody
> have any clues as to how to uncover the leak? I have a workaround to the
> browser behavior, but thats dancing around the issue, really. The
> browser does dot always behave like this, but I guess I can hack up a
> perl script to recreate the problem more reliantly if needed...
>
> I have attached (part of) a Ethereal summary to display the TCP/browser
> access pattern. The server is at 10.10.132.15, the client browser at
> 10.10.130.96. Needless to say, FF works like a champ...
>
If you think you can produce a small-ish test driver (perl?)
which can cause this, I think that would be the best way forward.
Then we can duplicate the problem and try to attack it.
--
------------------------------------------------------------
Gary Thomas | Consulting for the
MLB Associates | Embedded world
------------------------------------------------------------
--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
More information about the Ecos-discuss
mailing list