Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 10 Dec 2008 18:14:23 +0300
From: Eygene Ryabinkin <>
Subject: Re: CVE Request (nagios)

Andreas, thanks for answering.

Wed, Dec 10, 2008 at 03:53:47PM +0100, Andreas Ericsson wrote:
> >> So
> >>
> >> just completely closes the processing of these commands from the
> >> Nagios side.  May be this was the fix for the case when the evil
> >> contents from the command file were still floating around but the
> >> upgraded Nagios won't process them because they could go from the
> >> previous successful attack but are lying unprocessed?
> > 
> > Do you think it is really so?
> > 
> Umm... I can't parse the above paragraph.

I mean that in 3.0.6 even Nagios server won't execute CHANGE_
commands, because the diff in the above reference stops them from
being executed.

> In short though, the removed
> commands are removed *from the cgi's* because it's far too dangerous
> to allow such things over the web.

Comment in cgi.c says:
----- function cmd_submitf
         * We disallow sending 'CHANGE' commands from the cgi's
         * until we do proper session handling to prevent cross-site
         * request forgery
        if (!command || (strlen(command) > 6 && !memcmp("CHANGE", command, 6)))
                return ERROR;
So I presume that the danger comes from the CSRF.  This code was
introduced in 3.0.5.

> Nagios will still process them if
> they are submitted to the command-pipe, but the CGI's can no longer
> write such commands to said pipe.

Not in 3.0.6, see below and above.

> > CVE-2008-5028 really speaks about 3.0.5 as about vulnerable to CSRF.  At
> > least CHANGE_ commands were closed in 3.0.5 and were (presumably)
> > additionally closed at the Nagios server side in 3.0.6.  So either 3.0.6
> > is vulnerable too, 3.0.5 is not vulnerable to CSRF or I am missing
> > something.  What to choose?
> > 
> 3.0.5 is vulnerable to CSRF. 3.0.6 (which adds in-form session tokens to
> cmd.cgi, which processes all commands from the web-forms), is not vulnerable
> to CSRF.

If you're talking about the commit based on;a=commitdiff;h=9c2a418ab4f6e4ef3a53ddcde402fe4781caa764
then I afraid that this code isn't in the 3.0.6.  Diffing 3.0.5 and
3.0.6 yeilds some improvements, the hunk that Jan mentioned (it closes
CHANGE_ commands processing by the Nagios server itself):
--- nagios-3.0.5/base/commands.c        2008-11-02 21:51:29.000000000 +0300
+++ nagios-3.0.6/base/commands.c        2008-11-30 20:22:58.000000000 +0300
@@ -2891,6 +2893,19 @@
        unsigned long hattr=MODATTR_NONE;
        unsigned long sattr=MODATTR_NONE;

+       /* SECURITY PATCH - disable these for the time being */
+       switch(cmd){
+               return ERROR;
+               }
        /* get the command arguments */

> 3.0.5 fixes the authorization bypass discussed in CVE-2008-5027, where an
> authenticated user can submit commands he/she was not supposed to be able
> to submit.

Yes, newlines in the comments and other places.  This is really fixed
in 3.0.5.

> However, by blocking the CHANGE_ set of commands, the worst-case
> impact of the CSRF was drastically reduced, and the change to blocking those
> commands was also a part of 3.0.5.

Yes, I meant precisely this.  But again, no real CSRF fixes are present
in 3.0.6.

> I'm afraid Ethan (the Nagios maintainer) got it wrong in the changelog,
> which is why, I presume, there's so much confusion right now.
> I wrote the patches for it though, so I think it's safe to say I know what
> patch (and version) fixed what.

I understand this.  But I feel that you think of your session tokens
work as of being committed to 3.0.6.  This seems to be wrong.

Sorry for such a long letter.

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.