pcp
[Top] [All Lists]

Re: pmwebd security (was Re: [RFC] dynamic container switching)

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: pmwebd security (was Re: [RFC] dynamic container switching)
From: "Frank Ch. Eigler" <fche@xxxxxxxxxx>
Date: Tue, 10 Nov 2015 21:31:03 -0500
Cc: PCP <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <319465283.6999575.1447038990025.JavaMail.zimbra@xxxxxxxxxx>
References: <1313883527.54143616.1444783810135.JavaMail.zimbra@xxxxxxxxxx> <20151016223319.GH27211@xxxxxxxxxx> <1384643676.62705033.1445899239483.JavaMail.zimbra@xxxxxxxxxx> <20151027155234.GB9303@xxxxxxxxxx> <1185678657.63582036.1446001295613.JavaMail.zimbra@xxxxxxxxxx> <20151031022337.GC28852@xxxxxxxxxx> <1634759327.3957536.1446687727569.JavaMail.zimbra@xxxxxxxxxx> <20151106202554.GE2349@xxxxxxxxxx> <319465283.6999575.1447038990025.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Mutt/1.4.2.2i
Hi -

> > What?  No.  That CVE was about [...]
> 
> This is the summary, from the CVE:
> "Performance Co-Pilot (PCP) before 3.6.5 exports some of the /proc file
> system, which allows attackers to obtain sensitive information such as
> proc/pid/maps and command line arguments".

An informed person does not interpret a CVE solely based on its
summary line, which misses detail, context, severity.  (If one did,
one could fault /usr/bin/pminfo the same way.)


> [...] I'll make the needed pmwebd changes this week.

Saw the proposed changes in your git branch.  Restricting looks OK as
a default.  The local:* stuff should not be marked as deprecated nor
be undocumented: that still has plenty of safe & appropriate use.


> > pmwebd, like pmproxy, lacks detailed outgoing ACL facilities - see
> 
> Again, pmproxy does not create (automatically authenticated) unix: or
> local-context connections on behalf of remote clients, so it is not
> exposed to this problem.

An outgoing ACL is required to even contemplate putting pmwebd or
pmproxy onto a hostile network (e.g., to prevent their use as network
scanners), and would subsume the proposed pmwebd "-P" flag.


> > > > When a 1-line bash script can DoS pmcd  [...]
> > > What???  Could you please supply that script? - thanks.
> > [...]
> 
> You've forgotten to supply the "1-line bash script" ...?

OK, if you need me to spell out "pmcd now only hangs for a few seconds
after a hostile client sends a few bytes of data instead of
indefinitely" in code:

(printf "\x00\x00\x01" ; sleep 5) > /dev/tcp/localhost/44321

But, as I explained, there are multiple design limitations at play;
solving this one will get pmcd only a tiny bit closer to being
deployable directly on a hostile network.  (If you're serious about
it, expect to spend months.)


- FChE

<Prev in Thread] Current Thread [Next in Thread>