Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 11 Mar 2011 00:03:38 +0300
From: Solar Designer <>
To:, Florian Zumbiehl <>
Cc: Josh Bressers <>,
	"Steven M. Christey" <>,
	Stefan Fritsch <>, Petr Uzel <>,
	Thomas Biege <>, Jan Kalu??a <>
Subject: Re: CVE Request -- logrotate -- nine issues

Florian, all -

I'm sorry I failed to find time to comment on many postings in this
thread, and also in the thread about a possible vendor-sec alternative.

On Thu, Mar 10, 2011 at 07:08:38PM +0100, Florian Zumbiehl wrote:
> What about these?:
> | However, I think that still #6 (shell injection) and #7 (logrotate
> | DoS with strange characters in file names) should be considered
> | vulnerabilities in logrotate: It would be reasonable to assume that you
> | can use user input that's a valid (slash-less) filename as a (part of a)
> | log file name (assuming that the program is running as the same user that
> | inspects and rotates the logs, so the log directory being writable by
> | the program would not be insecure per-se) without that file name being
> | interpreted by a shell or causing logrotate to stop functioning,
> | respectively.

Based on the description above only (I have not looked at the code), it
appears that on one hand the program might behave unexpectedly, but on
the other no privilege boundary is crossed.

My understanding is that these issues would need to be treated as
vulnerabilities (and assigned CVE ids) only if they can be demonstrated
to have security impact (as in: someone manages to do more than they
were supposed to be able to) in an otherwise sane setup.

I guess this could be the case if logrotate config files, including log
filename patterns, are generated based on input from another user
(different than the service pseudo-user and not root) - e.g., via a
web-based administration interface with its own access control.
However, even if so (which already feels somewhat unrealistic) I have
difficulty imagining a web-based admin user who would be permitted to
configure log rotation (including filename patterns) yet would not
otherwise have access to the service pseudo-user account (perhaps the
person would be the server admin, including full root access).

To summarize, it feels like in theory a privilege boundary could exist
here and be crossed on certain systems with extra software, but in
practice this is unlikely and it would indicate poor design of another
piece of software or/and false sense of security put into that privilege
boundary.  I don't know what this means for CVE id assignment per the
current "rules".

Once again, I am relying on the description posted in here only (and
moreover on my interpretation of it), so maybe the actual situation is

I usually don't care much about CVE ids being assigned or not.  I cared
about that for the majority of these logrotate issues because it felt
like it could affect whether other packages will be fixed or not.  This
concern does not apply in the case of these two issues.  Thus, I have no
objections to CVE ids being assigned for them, even though we're not
going to treat them as vulnerabilities (based on info available so far).


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.