Security and dialbacks

[Dave Emery, Software]

In article <906 at hoptoad.uucp> gnu at hoptoad.UUCP writes:
>Here are the two messages I have archived on the subject...

>[I believe the definitive article in that discussion was by Lauren Weinstein,
>vortex!lauren; perhaps he has a copy.

	What follows is the original article that started the discussion.
I do not know whether it qualifies as the "definitive article"  as I
think I remember Lauren and I both posted further comments.
							    - Dave

		** ARTICLE FOLLOWS **

----------------------------------------------------------------------

	An increasingly popular technique for protecting dial-in ports from
the ravages of hackers and other more sinister system penetrators is dial
back operation wherein a legitimate user initiates a call to the system
he desires to connect with, types in his user ID and perhaps a password,
disconnects and waits for the system to call him back at a prearranged number.
It is assumed that a penetrator will not be able to specify the dial back
number (which is carefully protected), and so even if he is able to guess
a user-name/password pair he cannot penetrate the system because he cannot
do anything meaningful except type in a user-name and password when he is
connected to the system. If he has a correct pair it is assumed the worst that
could happen is a spurious call to some legitimate user which will do no harm
and might even result in a security investigation.

	Many installations depend on dial-back operation of modems for
their principle protection against penetration via their dial up ports
on the incorrect presumption that there is no way a penetrator could
get connected to the modem on the call back call unless he was able to
tap directly into the line being called back.  Alas, this assumption
is not always true - compromises in the design of modems and the
telephone network unfortunately make it all too possible for a clever
penetrator to get connected to the call back call and fool the modem 
into thinking that it had in fact dialed the legitimate user.

	The problem areas are as follows:

		Caller control central offices

	Many older telephone central office switches implement caller
control in which the release of the connection from a calling telephone
to a called telephone is exclusively controlled by the originating
telephone.  This means that if the penetrator simply failed to hang up
a call to a modem on such a central office after he typed the legitimate
user's user-name and password, the modem would be unable to hang up the
connection.

	Almost all modems would simply go on-hook in this situation
and not notice that the connection had not been broken.  If the same line
was used to dial out on as the call came in on,  when the modem
went to dial out to call the legitimate user back the it might not
notice (there is no standard way of doing so electrically) that the
penetrator was still connected on the line.  This means that the modem
might attempt to dial and then wait for an answerback tone from the far 
end modem. If the penetrator was kind enough to supply the answerback tone
from his modem after he heard the system modem dial, he could make a connection
and penetrate the system. Of course aome modems incorporate dial tone detectors
and ringback detectors and in fact wait for dial tone before dialing, and
ringback after dialing but fooling those with a recording of dial tone 
(or a dial tone generator chip) should pose little problem.

	
		Trying to call out on a ringing line

	Some modems are dumb enough to pick up a ringing line and
attempt to make a call out on it.   This fact could be used by a
system penetrator to break dial back security even on joint control or
called party control central offices.  A penetrator would merely have to
dial in on the dial-out line (which would work even if it was a separate
line as long as the penetrator was able to obtain it's number), just as
the modem was about to dial out.  The same technique of waiting for
dialing to complete and then supplying answerback tone could be used - and
of course the same technique of supplying dial tone to a modem which waited
for it would work here too.

	Calling the dial-out line would work especially well in cases where the
software controlling the modem either disabled auto-answer during the period
between dial-in and dial-back (and thus allowed the line to ring with no action
being taken) or allowed the modem to answer the line (auto-answer enabled) and
paid no attention to whether the line was already connected when it
tried to dial out on it.


		The ring window

	However, even carefully written software can be
fooled by the ring window problem.  Many central offices actually will connect
an incoming call to a line if the line goes off hook just as the call comes
in without first having put the 20 hz. ringing voltage on the line to make it
ring.  The ring voltage in many telephone central offices is supplied
asynchronously every 6 seconds to every line on which there is an incoming call
that has not been answered, so if an incoming call reaches
a line just an instant after the end of the ring period and the line
clairvointly responds by going off hook it may never see any ring voltage.

	This means that a modem that picks up the line to dial out just as our
penetrator dials in may not see any ring voltage and may therefore have no
way of knowing that it is connected to an incoming call rather than
the call originating circuitry of the switch.  And even if the switch
always rings before connecting an incoming call, most modems have a
window just as they are going off hook to originate a call when they
will ignore transients (such as ringing voltage) on the assumption that
they originate from the going-off-hook process. [The author is aware
that some central offices reverse battery (the polarity of the voltage
on the line) in the answer condition to distinguish it from the
originate condition, but as this is by no means universal few if any
modems take advantage of the information supplied] 


		In Summary

	It is thus impossible to say with any certainty that when a modem
goes off hook and tries to dial out on a line which can accept incoming calls
it really is connected to the switch and actually making an outgoing call.
And because it is relatively easy for a system penetrator to fool the
tone detecting circuitry in a modem into believing that it is seeing dial
tone, ringback and so forth until he supplies answerback tone and connects
and penetrates system security should not depend on this sort of dial-back.


		Some Recommendations

	Dial back using the same line used to dial in is not very secure
and cannot be made completely secure with conventional modems.  Use of dithered
(random) time delays between dial in and dial back combined with allowing
the modem to answer during the wait period (with provisions made for
recognizing the fact that this wasn't the originated call - perhaps by
checking to see if the modem is in originate or answer mode) will
substantially reduce this window of vulnerability but nothing can completely
eliminate it.

	Obviously if one happens to be connected to an older caller control
switch, using the same line for dial in and dial out isn't secure at 
all.  It is easy to experimentally determine this, so it ought to be possible
to avoid such situations.

	Dial back using a separate line (or line and modem) for dialing
out is much better, provided that either the dial out line is sterile
(not readily tracable by a penetrator to the target system) or that it is a one
way line that cannot accept incoming calls at all.  Unfortunately the later
technique is far superior to the former in most organizations as concealing
the telephone number of dial out lines for long periods involves considerable
risk.  The author has not tried to order a dial out only telephone line,
so he is unaware of what special charges might be made for this service
or even if it is available.

		A final word of warning

	In years past it was possible to access telephone company test
and verification trunks in some areas of the country by using mf tones from so
called "blue boxes". These test trunks connect to special ports on telephone
switches that allow a test connection to be made to a line that doesn't
disconnect when the line hangs up.   These test connections could
be used to fool a dial out modem, even one on a dial out only line (since
the telephone company needs a way to test it, they usually supply test
connections to it even if the customer can't receive calls).

	Access to verification and test ports and trunks has been tightened
(they are a kind of dial-a-wiretap so it ought to be pretty difficult)
but in any as in any system there is always the danger that someone, through
stupidity or ignorance if not mendacity will allow a system penetrator
access to one.

		**  Some more recent comments **

	Since posting this I have had several people suggest use
of PBX lines that can dial out but not be dialed into or outward WATS
lines that also cannot be dialed.  Several people have also suggested
use of call forwarding to forward incoming calls on the dial out
line to the security office. [ This may not work too well in areas
served by certain ESS's which ring the number from which calls are
being forwarded once anyway in case someone forgot to cancel forwarding.
Forwarding is also subject to being cancelled at random times by central
office software reboots]

	And since posting this I actually tried making some measurements
of how wide the incoming call window is for the modems we use for dial
in at CRDS.  It appears to be at least 2-3 seconds for US Robotics
Courier 2400 baud modems.  I found I could defeat same-line-for-dial-out  
dialback quite handily in a few dozen tries no matter what tricks I
played with timing and watching modem status in the dial back login software.
I eventually concluded that short of reprogramming the micro in the modem
to be smarter about monitoring line state, there was little I could do at
the login (getty) level to provide much security for same line dialback.

	Since it usually took a few tries to break in, it is possible to
provide some slight security improvement by sharply limiting the number of
unsucessful callbacks per user per day so that a hacker with only
a couple of passwords would have to try over a significant period of time.

	Note that dialback on a dedicated dial-out only line is 
somewhat secure.


          David I. Emery    Charles River Data Systems   617-626-1102
          983 Concord St., Framingham, MA 01701.
	  uucp: decvax!frog!die

-- 
          David I. Emery   Charles River Data Systems
983 Concord St., Framingham, MA 01701 (617) 626-1102 uucp: decvax!frog!die