Vulnerability6 min read

Blaster: the worm that hit 100,000 Windows PCs in a week

In August 2003, the Blaster worm infected more than 100,000 Windows computers in its first week. How it spread, why it attacked Microsoft and why it stayed.

Illustration: a ring showing the CVSS score 7.5, with the title Blaster

What to do now

  1. On any computer that still runs Windows NT 4.0, 2000, XP or Server 2003, install Microsoft’s update MS03-039 (KB824146): it contains the fix from MS03-026 and closes three more flaws in the same service.
  2. If you cannot install it at once, switch off DCOM as Microsoft describes in article KB825750; programs that use DCOM across the network then stop working.
  3. Block TCP ports 135, 139, 445 and 593 and UDP ports 135, 137, 138 and 445 at your firewall, so that no one can reach RPC from the internet.
  4. Replace these old Windows versions: they no longer receive regular security updates.
Year
2003
CVSS 2.0 score
7.5 high
Windows computers infected in the first week
100,000+
doubling time of the number of infected computers in the first hours
9 min
Why it is on the list
  • Reach met

    Blaster infected more than 100,000 Windows computers in its first week; on its busiest day, a monitored address block covering 1/256 of the internet saw about 106,000 infected addresses (The Blaster Worm: Then and Now, 2005).

  • Mass exploitation met

    Blaster was a self-spreading worm: every infected computer searched for new victims and infected them on its own (CERT/CC; The Blaster Worm: Then and Now, 2005).

  • Damage not met

    The sources give no total of the damage and document no outages of critical services; a survey of 19 research universities found that each spent an average of USD 299,579 to recover (The Blaster Worm: Then and Now, 2005).

  • Extraordinary response met

    Microsoft provided its fix even for Windows versions whose support had ended, took the attacked address windowsupdate.com out of the DNS and, joined by the FBI, the US Secret Service and Interpol, offered USD 250,000 for the authors (Microsoft; SANS Internet Storm Center; US Secret Service).

How a vulnerability gets on this list

The record
AffectedWindows NT 4.0, 2000, XP and Server 2003, through the RPC service that runs by default
CVECVE-2003-0352
CVSS vectorAV:N/AC:L/Au:N/C:P/I:P/A:P
Weakness (CWE)NVD-CWE-Other
Fixed inMicrosoft patch MS03-026, KB823980 (July 16, 2003); replaced by MS03-039, KB824146 (September 10, 2003)
Found byThe Last Stage of Delirium Research Group
Timeline
12345678
  1. 1Fix available
  2. 2Working attack code appears on a public mailing list+10 days
  3. 3First attacks seen+15 days
  4. 4Outbreak begins+26 days
  5. 5Microsoft takes windowsupdate.com, the target of the worm's attack, out of the DNS+30 days
  6. 6Welchia, also called Nachi, a second worm, uses the same flaw+33 days
  7. 7Microsoft fixes three more flaws in the same service with MS03-039+56 days
  8. 8Microsoft, joined by the FBI, the US Secret Service and Interpol, offers USD 250,000 for the authors+112 days

On August 11, 2003, a worm called Blaster, also known as MSBlast or Lovsan, began to spread through a flaw in Windows that Microsoft had fixed four weeks earlier. A worm is a program that spreads from computer to computer on its own. Blaster needed no click and no password: an unpatched Windows computer within its reach was enough. In its first week, it infected more than 100,000 Windows computers, according to a measurement study, and a year later it was still spreading.

How it worked

Windows uses a service called RPC, short for Remote Procedure Call, to let programs call functions on other computers. One part of it handles DCOM, a way for software components to work together across a network. According to Microsoft, the service runs by default in all versions of Windows. It listens on several network ports, numbered doors for network traffic, above all TCP port 135.

One DCOM function did not properly check incoming messages. A crafted message overflowed a memory area of fixed size, a buffer, and could run the attacker’s code with Local System privileges, the highest rights in Windows. Windows NT 4.0, 2000, XP and Server 2003 were all affected.

Blaster reused attack code that had been published on a public mailing list in July. Each infected computer picked a starting address, in 40 % of cases one close to its own, and tried 20 addresses at a time on port 135. A successful attack opened a backdoor, a hidden command line, on port 4444. Through it, the new victim was told to fetch the worm file, msblast.exe, from the attacking computer with TFTP, a simple file transfer method, and to run it. An entry in the registry, the central settings database of Windows, restarted the worm after every reboot.

Microsoft warned that Blaster could make infected PCs crash again and again. CERT/CC, the US computer emergency response center at Carnegie Mellon University, reported that some machines could not stay online long enough to download the patch. The worm also had a second task: on August 16, every infected computer was to flood windowsupdate.com, an address of Microsoft’s update website, with connection requests.

  1. Infected PCtries 20 addresses at a time on port 135
  2. RPC servicea crafted DCOM message overflows a buffer
  3. Backdooropens a command line on port 4444
  4. New victimdownloads msblast.exe by TFTP and runs it
No click and no password were needed. Each new victim started the same steps on its own.

Reach and damage

The best measurements come from “The Blaster Worm: Then and Now”, a 2005 study by researchers at the University of Michigan and the security firm Arbor Networks. They watched a block of about 16 million unused internet addresses, one 256th of the internet. No real computers use them, so any traffic arriving there comes from scans, such as worms looking for victims.

After Microsoft’s warning in July, the number of computers probing port 135 on this block rose from 4–10 a day to 100–300 a day. When Blaster struck, the number of infected computers at first doubled about every nine minutes. At the peak, the researchers saw nearly 15,000 infected addresses in one hour and about 106,000 in 24 hours. After eight hours, activity began to fall, with a half-life of about 12 hours, as owners patched or disconnected computers and networks blocked the worm.

The infected networks included several consumer broadband providers. Other counts differ widely:

Three ways of counting Blaster
Monitored address block, busiest day106,000
DShield firewall logs, highest count208,000
Microsoft cleanup tool, users helped8,000,000

The first two bars count internet addresses that probed port 135, the third the users whom Microsoft’s cleanup tool helped, by Microsoft’s count. Sources: The Blaster Worm: Then and Now (2005); Microsoft (2004).

DShield, a service that collects firewall logs from tens of thousands of devices, saw at most 208,000 addresses scanning port 135, against about 7,000 before the worm. In February 2004, Microsoft said that its cleanup tool had helped more than 8 million internet users; the researchers note that this differs widely from both measurements.

No reliable total of the damage exists. A survey of 19 research universities, quoted in the study, found that each spent an average of USD 299,579 over five weeks to recover from Blaster and its variants.

On August 18, a second worm began to use the same flaw. Welchia, also called Nachi, was meant as an “antiworm”: according to the study, it tried to patch the flaw on the computers it infected. Before each attack, it sent a ping, a small test message, to see whether a computer was online. The network equipment maker Cisco reported that the high volumes of these packets caused many of the network problems, including overloaded routers. The study concludes that Welchia caused significant damage of its own.

The fix and the response

The research group The Last Stage of Delirium had reported the flaw to Microsoft. On July 16, 2003, Microsoft published security bulletin MS03-026 with a patch, rated the flaw critical for every affected version of Windows and advised: “Systems administrators should apply the patch immediately.” As a stopgap, it suggested blocking the RPC ports at the firewall or switching off DCOM. No public attack code existed yet. By July 26, working attack code was on a public mailing list, and on July 31, CERT/CC reported widespread scanning and exploitation.

After the outbreak, Microsoft went beyond its usual practice and provided its patch for Windows 2000 Service Pack 2 and Windows NT 4.0 Workstation as well, although their support had already ended. On September 10, 2003, its bulletin MS03-039 fixed three more flaws in the same service and replaced MS03-026; for the old versions, Microsoft made an exception “due to the nature of this vulnerability”.

Microsoft also defused the attack on windowsupdate.com. On August 15, the day before, it stopped resolving that address in the DNS, the address book of the internet, according to the SANS Internet Storm Center, which tracks internet attacks. Infected computers found nothing to attack, while Windows’ automatic updates, which used windowsupdate.microsoft.com, kept working.

On November 5, 2003, Microsoft, joined by the FBI, the US Secret Service and Interpol, offered USD 250,000 for information leading to the arrest and conviction of those behind the original worm. People behind two later variants had already been arrested; the author of the original was never caught, the study notes. Millions downloaded Microsoft’s cleanup tool in January 2004, but according to the study, the number of infections barely changed.

A year later, in August and September 2004, the researchers counted more than 200,000 different addresses still spreading the worm, up to 4,100 a day. Most were new: 99.5 % of the addresses, and 73 % of the surrounding blocks of 256 addresses, had not appeared during the outbreak. The researchers conclude that the worm moved on to new computers as old ones were cleaned or switched off.

Infected addresses seen on the busiest day
August 2003106,000
August–September 20044,100

Addresses that scanned the monitored block within one day, during the outbreak and one year later. Source: The Blaster Worm: Then and Now (2005).

The National Vulnerability Database rates the flaw 7.5 out of 10 on the older CVSS version 2, with no score on version 3. That score assumes only partial harm to confidentiality, integrity and availability, although Microsoft says an attacker could gain complete control of the computer.