Re: Varnish - no CVE == bug regression

[Stefan Bühler <stbuehler () lighttpd net>]

On Thu, 3 Jul 2014 08:15:06 +0000
Sven Kieske <S.Kieske () mittwald de> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> I'd agree with this.
> And I don't get the argument from poul-henning kamp, what I understand
> is:
> "hey, we trust our backend server"
> well, but your backend server can make you crash, so you probably
> shouldn't trust it in the first place?
>
> you _never_ can trust input, so you have to validate it, either way,
> at least enough to not crash or perform malicious actions.
>
> Am 03.07.2014 09:48, schrieb Kurt Seifried:
> > So as I understand this: Varnish front end for web servers, the web
> > servers can trigger varnish to restart. Are the back end servers
> > supposed to be able to cause varnish to restart?
> >
> > I'm guessing not. Scenario: hosting env, or a website with a vuln,
> >  whatever, you can now cause the varnish front ends to restart 
> > constantly, effectively causing a permanent denial of service.
> >
> > That sounds CVE worthy. Or am I missing something?

you should never trust *untrusted* input. your root shell usually
trusts the input it gets...

so the valgrind developers decided that they consider the backend
webservers trusted, at least regarding the capability to cause a DoS.

for the record - so does lighttpd (a backend can trigger OOM as lighty
reads (nearly) as fast as possible from a backend, as backends often
only handle one request at a time); we usually tell people to use
X-sendfile instead of sending ISOs through php.

just because you disagree with such decisions doesn't make it CVE
worthy (missing or wrong documentation could).

in case you actually want to assign a CVE here, maybe we can get one
for the bad openssl default cipherstring too? because for that it is
really obvious that it is f*** wrong, but i think that none was
assigned because upstream didn't agree with it.

regards,
Stefan