[wilhelmtux-discussion] m$ wurm sabotiert internet

Mathias Gygax mg at trash.net
Sam Jan 25 12:09:28 CET 2003

[ wie heute morgen (lokale zeit) auf der nanog mailing liste diskutiert
  wurde, schnellen momentan die lasten der internet core router massiv
  an. es handelt sich dabei um einen wurm der m$ sql server infisziert
  und UDP floods in alle netze schickt. wer informiert ist, kennt die
  attacke.  verbindet man via ip source spoofing zwei echo dienste
  miteinander flooden die einander solange (ohne das der angreifer ein
  weiteres packet schicken müsste) bis sie abgeschaltet werden. dasselbe
  passiert hier mit den m$ sql servern. server 1 schiesst server 2 ab
  und umgekehrt. massiv skaliert haben wir den dämlichsten, aber
  effektivsten wurm seit langem.

  go home, bill! :o)



NGSSoftware Insight Security Research Advisory
Name: Unauthenticated Remote Compromise in MS SQL Server 2000
Systems: Microsoft SQL Server 2000, all Service Packs
Severity: Critical/Very High Risk.
Category: Remote Buffer Overrun Vulnerability
Vendor URL: http://www.microsoft.com/
Author: David Litchfield (david at ngssoftware.com)
Advisory URL: http://www.ngssoftware.com/advisories/mssql-udp.txt
Date: 25th July 2002

Advisory number: #NISR25072002

VNA reference : http://www.ngssoftware.com/vna/ms-sql.txt

This advisory covers the solution to one of the problems mentioned in the
above VNA URL.


Microsoft's database server SQL Server 2000 exhibits two buffer overrun
vulnerabilities that can be exploited by a remote attacker without ever
having to authenticate to the server. What further exacerbates these issues
is that the attack is channeled over UDP. Whether the SQL Server process
runs in the security context of a domain user or the local SYSTEM account,
successful exploitation of these security holes will mean a total compromise
of the target server and its data.


SQL Server can be configured to listen for incoming client connections in
several different ways. It can be configured such that clients can use named
pipes over a NetBIOS session (TCP port 139/445) or sockets with clients
connecting to TCP port 1433 or both. Which ever method is used the SQL
Server will always listen on UDP port 1434. This port is designated as the
Microsoft SQL Monitor port and clients will send a message to this port to
dynamically discover how the client should connect to the Server. This
message is a single byte packet, the byte being 0x02.

There are other messages that can be sent to this port and these can be
worked out with simple experimentation.

Stack Based Buffer Overflow

When SQL Server receives a packet on UDP port 1434 with the first byte set
to 0x04, the SQL Monitor thread takes the remaining data in the packet and
attempts to open a registry key using this user supplied information. For
example, by sending \x04\x41\x41\x41\x41 (0x04 followed by 4 upper case
'A's) SQL Server attempts to open

HKLM\Software\Microsoft\Microsoft SQL Server\AAAA\MSSQLServer\CurrentVersion

By appending a large number of bytes to the end of this packet, whilst
preparing the string for the registry key to open, a stack based buffer is
overflowed and the saved return address is overwritten. This allows an
attacker to gain complete control of the SQL Server process and its path of
execution. By overwriting the saved return address on the stack with an
address that contains a "jmp esp" or "call esp" instruction, when the
vulnerable procedure returns the processor will start executing code of the
attacker's choice. At no stage does the attacker need to authenticate.

Heap Based Buffer Overflow

When SQL Server receives a packet on UDP port 1434 with the first byte set
to 0x08 followed by an overly long string, followed by a colon character (:)
and number a heap based buffer is overflowed. As this corrupts the
structures used to keep track of the heap an attacker can overwrite any
location in memory with 4 bytes of their own choosing. This can be used to
gain remote control of the processes execution. If the colon and number are
missing the SQL Server process access violates before the heap is corrupted
as the code in the SQL Monitor thread fails to handle exceptions.

For example the code calls the C function strtok(). The strtok() functions
looks for a given token in a string, in this case a colon, and if found
returns a pointer to it. If the colon is missing in the string being
searched then no pointer is returned. This is one of the reasons why the SQL
Server process access violates if the colon is missing. The code does not
check to see if a valid pointer has been returned before passing it to
another function call, atoi():

char *ptr=NULL;
int num=0;
ptr = strtok(string,":");
num = atoi(ptr); // ptr is used without being validated

Failure to check return values and handle exceptions leads to the process
dying, leading to a simple Denial of Service attack. That said, in the light
of the overflows, the DoS is the least of the problems.

Network Based Denial of Service

When an SQL Server receives a single byte packet, 0x0A, on UDP port 1434 it
will reply to the sender with 0x0A. A problem arises as SQL Server will
respond, sending a 'ping' response to the source IP address and source port.
This 'ping' is a single byte UDP packet - 0x0A. By spoofing a packet from
one SQL Server, setting the UDP port to 1434, and sending it the a second
SQL Server, the second will respond to the first's UDP port 1434. The first
will then reply to the second's UDP port 1434 and so on. This causes a storm
of single byte pings between the two servers. Only when one of the servers
is disconnected from the network or its SQL service is stopped will the
storm stop. This is a simple newtork based DoS, reminiscent of the echo and
chargen DoSes discussed back in 1996
(http://www.cert.org/advisories/CA-1996-01.html). When in this state, the
load on each SQL Server is raised to c. 40 - 60 % CPU time.

Considerations for protection against these vulnerabilities

Exploitation of these security holes goes over UDP, a connection-less
communications protocol. As such it makes the task of bypassing the
protection offered by a firewall considerably easier. The spoofing of an IP
address in a UDP packet is also considerably easier.

It is trivial for an attacker to send an attack through the firewall,
setting the source IP address to that of the target's DNS Server and the
source port to 53. Most firewalls will allow this packet through as it will
look like a response to a query to resolve a domain name.

It is strongly recommended that a rule be added to each organization's
firewall such that any packet destined for UDP port 1434 on the 'clean' side
of the firewall be dropped and logged. No host, even DNS Servers, should be
allowed to send traffic to this port.

It is also recommend that firewall administrators ensure that any packet
received on the 'dirty' interface with a source IP address set to an address
on the clean side is also dropped and logged.

Fix Information

NGSSoftware alerted Microsoft to this problem on the 17th of May 2002 and
they have produced a patch that resolves these issues. NGSSoftware urge all
customers of SQL Server 2000 to test then apply these fixes as soon as


Where possible, NGSSoftware also recommend running the SQL Server as low
privileged local account and not SYSTEM or a domain account.

A check for these vulnerabilities has been added to Typhon II,
NGSSoftware'svulnerability assessment scanner, of which, more information is
available from the NGSSite, http://www.ngssoftware.com/

Further Information
The author will be discussing these vulnerabilities in detail, with several
exploitation demonstrations, at the up and coming Blackhat Security
Briefings conference is Las Vegas. For those interested in attending this
conference, please see http://www.blackhat.com for more details.

For more information about buffer overflows please read