copyright notice
link to published version: Communications of the ACM, December, 2005

accesses since September 23, 2005


Hal Berghel

David Hoelzer

It is now well documented that some of the greatest security vulnerabilities to computer software are a result of inattention to current service packs, hot fixes and patches. Our experience with big-ticket malware such as W32/Blaster and SoBig illustrate the point: in both cases patches that would have prevented the infections were available at least a month before the malware was deployed (cf. this column in the December, 2003 CACM). The successes of these exploits were a consequence of the millions of computers that hadn't been patched, either through neglect or ignorance.

This leads us to a study of the art of preemption: taking preventative measures against potential insecurities. This is indeed an art rather than a science, for the problem space is enormous - the Cartesian cross-product of

(a) 131,072 available ports (65,536 each for TCP and UDP),
(b) hundreds of well-known network services (e.g.,telnet, ftp),
(c) thousands of network applications (both standard and non-standard), and
(d) an ever-increasing fleet of network-mobile trojans, viruses, worms, and rootkits.

Even this analysis is over-simplified. Things are actually much worse when one adds in all of the present and future malware for which no betraying signature or operational characteristics have as yet been discovered. For example, zero-day exploits, by their very nature, are not taken account of in (d). This problem deserves consideration regardless of our network security precautions (firewalls, etc.) which has been well demonstrated by worms like SoBig, NetSky, Bagel, etc.

Our present discussion will focus on (a) - Internet Ports, part of the problem space.

TCP/IP and Ports

The internet operates on the 65,536 TCP and 65,536 UDP ports. The official IANA ( classification divides these ports into three categories: well-known ports (0-1023), registered ports (1024-49151), and dynamic and/or private ports (49152-65535), with many pockets of reserved ports dispersed throughout. In general port assignments are a result of an approval process between a petitioner and IANA. However, some caveats are in order:

  1.   IANA approves ports based on the application and intended use. There is no warranty expressed or implied by IANA that using a port is a good idea: "Assignment of a port number does not in any way imply an endorsement of an application or product, and the fact that network traffic is flowing to or from a registered port does not mean that it is "good" traffic." ( Caveat emptor is the operational metaphor. An example of an ill-advised port might be 69 (trivial file transfer protocol). This is an exemplar of high-risk internetworking because of total absence of security features.
  2. There is little control over the actual use once the port is registered. The fact that IANA approved a port neither implies that (a) that is the only use to which the port is put, nor (b) the port may be safely used. For example, IANA approved port 1999 as a Cisco identification port. However, in practice it is one of the ports used by the infamous SubSeven backdoor and Trojan.
  3. There is also little control over the use of ports that aren't registered. In addition to registered ports 1999 and 2773 (RBackup), for example, SubSeven also uses unassigned ports (7215 and 27374), and at least one port in the dynamic/private range (54283)

The relevance of IANA registration has declined over time as more Internet vendors wandered off on their own. At this writing any rhyme or reason there might be behind port assignments is far from obvious. In the discussion to follow it is worth remembering that the Internet was built without consideration to security and authentication. Therein lies the rub.


For convenience, we group common port vulnerabilities as (1) Windows-centric, (2) Unix-centric, and (3) operating systems-independent, with considerable overlap recognized. This overview will describe a few of the better-known, port-related vulnerabilities that you're likely to encounter.

(1) Windows-Centric

Microsoft in general, and the Windows operating system in particular, receives a continuous beating in the media over security vulnerabilities - some deserved, some not. There is so much hostility in some of these attacks that it is sometimes difficult to remember that most of the problems result from Microsoft's commitment for maximum usability of their products. The trade-off between usability and security rears its ugly head whenever a software company seeks hegemony in its market. Nowhere was this more evident than in pre-XP Windows releases that came with virtually every feature enabled.

To illustrate, from a Windows XP command prompt enter <netstat - ano>. If one combines this information that the Windows Task Manager (TASKMAN: view>select columns>check "PID">OK) provides concerning the process id, one gets a pretty good idea of the active ports and services. For example, in Figure 1 wee see that process 1504, which TASKMAN identifies as <alg.exe>, is listening on TCP port 1025 via this computer's loopback address. For most of us questions like these arise: (a) exactly what does <alg.exe> do?, and (b) why does it need an open TCP connection to do it? This illustrates one of the greater security issues in Windows - most users are unaware THAT a large number of services are running much less WHY.

The Microsoft Vista operating system that is currently in Beta testing and is scheduled for release in late 2006 will address this issue by taking a Unix-like load-time minimalism. But at least for the moment, most of the responsibility for blocking unwanted services and Internet accesses accrues to the user. Toward that end, we need some sort of "best practices" advice on which ports to close and which to leave open. This section offers a modest beginning.


Figure 1: A Typical Windows Netstat Report of Active Ports and Services


TCP and UDP ports 135, 137-139 and 445 are particularly problematic in Windows. Technically, ports 135 and 137-139 are used by the legacy NetBios application programming interface while port 445 is assigned to a network protocol called Server Message Block. While originally intended for file and printing sharing among a cluster of computers in small workgroups, both may be deployed over TCP/IP as a distributed network file system. NetBios and SMB didn't scale to the Internet cloud gracefully. Microsoft used NetBios and SMB to support interprocess communication between and among the "network neighborhood" resources by using the SYSTEM ID. Since one can't log on as SYSTEM as it has no username or password the communication avoids both - hence the term "null session." The command to establish a null session on a local machine confirms the null usernames and passwords: net use \\\IPC$ "" /user: "" .

Null sessions have issues. For one, a fairly substantial body of malware uses these ports (e.g., Nimda, W32.Blaster,W32/Lovsan.worm, Bugbear, Chode, Qaz on the NetBios ports, and Nimda, Backdoor.rtkit.b, Sasser, W32.HLLW.Deloder, and several W32.spybot variations on SMB, to name but a few). For another, there are freely available tools available that exploit null sessions. Enum ( - not to be confused with the public switched telephone network - is one such utility. Enum establishes null sessions across networks and enumerates shares. It can also integrate with a lexicon to run a brute force attack on passwords and usernames. Another utility, DumpSec ( is a GUI-based resource enumerator. Figure 2 illustrates how simply DumpSec may be used to enumerate properties of user accounts. Beyond null session vulnerability, NetBios and/or SMB are also susceptible to a critical remote code execution vulnerability and a License Logging Service Overflow, so they fall short of the rock-solid mark on most scales.

While NetBios and SMB can be shut off in the Windows Registry, and can be unbound from the network interface as well, it's wisest to also block the ports at the firewall as an additional precaution. Bear in mind that the greatest vulnerability of NetBios and SMB is through Internet-access not from controlled internal LANs.

Further, these Remote Procedure Call vulnerabilities now extend beyond port 135. Since RPC is set up to bind to the first available TCP or UDP port above 1025, RPC vulnerabilities extend to these ports as well. This has become exascerbated with the popularity of Windows Messenger. LURHQ ( observed that large blocks of IP addresses recently began targeting ports 1026-9 with anonymous pop-up spam in response to ISP filtering port 135. Independent confirmation has been reported by SANS ( In addition, these ports are vulnerable to buffer overflow attacks before XP SP2. If you don't need Windows Messenger, closing ports 1026-9 is well advised.

Figure 2. DumpSec's Enumeration of User Account Properties.

Beyond the direct vulnerabilities, NetBios has also been used by media harvesters such as Scour . Though bankrupted in 2000 after being sued for copyright infringement, copies of the software remain on the Internet.

Another feature related to NetBios is the Windows Internet Naming Service (WINS) that operates on port TCP/UDP port 42. WINS handles name resolution on Windows computers so that the translation from NetBios names to IP addresses are stored internally on Windows computers that share resources. The vulnerability arises from the WINS replication service whereby Windows computers share these translation tables. This service is vulnerable to packet-crafting attacks that substitute a corrupted version of the address table to be substituted for the legitimate version and thereby misdirecting traffic to hacker-controlled computers (cf. Blocking TCP/UDP port 42 at the firewall avoids this vulnerability.

TCP/UDP Port 3389 is also worrisome. 3389 is the Windows Remote Desktop Protocol port. This protocol provides remote display and input capabilities for Windows applications running on servers. Though the capabilities are limited to SVGA display resolution and clipboard data transfer (not to mention being somewhat slow), features like printer redirection, virtual channels, support for remote system administration and technical support, and client-server display toggling make RPC a popular choice for remote connections. We note that with the exception of the Media Center Edition of XP, RDP is disabled by default.

The problem with WTS is that RDP has been shown to be vulnerable to denial-of-service attacks by malformed packets up to and including Service Pack 2 of Windows XP (cf. ). Though Microsoft announced both a security advisory and patch in July, 2005, unpatched computers remain vulnerable. Because of the nature of this vulnerability, and the fact that RDP has been beset by vulnerability reports for several years (cf., it is recommended that port 3389 be blocked at the firewall and that RDP be disabled as well.

(2)  Unix-Centric

Most UNIX systems, regardless of brand, offer SMTP mail services through TCP port 25. If the system is not actually a mail relay or server, then you likely do not need this service running at all. The main exposure lies in the failure to continuously patch the system. By the way, it is a common misconception that an SMTP server must be running for a local user or service to send email out of a UNIX box. This is simply not true. The recommended solution is to block TCP 25 at the firewall.

Ports 135-139, both TCP and UDP, have become extremely common on Unix machines, allowing them to interact with their Windows counterparts using the Samba service. While this is an excellent facility to allow for file and printer sharing, or even authentication services, it is also very common to find that the services permit absolutely anyone on the network to connect to the server. This can lead to the same sorts of exposures created by unprotected shares on Windows systems (see above), and can also be used to perform brute force authentication attempts or password guessing against the service to determine user names and passwords. The exposure is further amplified by the fact that the user names and passwords are often synchronized between the UNIX and Windows systems in order to allow for seamless integration. This means that once one system is compromised, all systems with that same user are also compromised. Ports 135-9 should be closed unless these services are critical.

TCP port 513 is typically used for the Rlogin service. This service, along with RSH and RCP, are considered ill-advised for use in public environments. Most security practitioners would also recommend that they be eliminated on private networks as well in favor of a service like SSHv.2, which can be used to perform all of the same "R" utility functions in a more secure manner (namely encrypted, possibly stronger authentication). However, the fact is that many college and university systems have these services enabled and may be actively using these services for convenience. If these services are offered, there must be some independent network level control in place since these services in particular make their authentication decisions based on the source address from which a connection arrives. Fooling these services is generally considered to be a trivial exercise. Adding network-level access controls at routers, firewalls and switches to limit communication with servers via TCP 513 is recommended.

TCP port 22 is used by the Secure Shell service. It may be surprising to find that a service designed for security is listed in our set of ports to watch carefully! The fact is that this service is a great target for attackers. Not only do we have the same vulnerabilities that exist with all ports (i.e., buffer overflows, heap overflows, etc.) when the patch level is not maintained, but the attacker also has the potential to create an encrypted session, thereby preventing monitoring systems from performing signature alerting on the contents of the traffic. Another substantial risk that exists here is the potential for brute forcing credentials. SSH systems are commonly configured to perform username/password authentication. This means that determining a username for a system (possibly from an email address), the attacker can now try brute force login attempts by bypassing the lockout restrictions that normally apply, since SSH is usually not integrated into the typical user authentication. A great way to limit this exposure is to either limit access to port 22 or use certificate based authentication (or both!).

TCP port 389 is typically used by the LDAP service. The purpose of this service is to provide directory services to allow lookups of names, email address, public keys, etc. Depending on what sort of information is stored in the LDAP server and how it is secured, it may be possible for someone to browse or query the LDAP server to recover usernames to be used in brute force guessing attempts or email addresses to fill up your mailbox with UCE (commonly known as spam). Blocking TCP and UDP port 389 is recommended, along with related LDAP ports 3268, 3269 and 636 especially in Windows environments.

Telnet has been around for years. It is extremely common to find this service in frequent use on internal and university networks. Since Telnet is a plain text protocol, all usernames and passwords are readable off the wire. SSHv2 is considered to be the optimal way to remotely connect to a system - even if the network is internal. There is a popular misconception that a switched network prevents someone from intercepting telnet communication with a sniffer. This is not true. Tools like Ettercap and DSniff, with ARP spoofing capabilities make this possible. Telnet ports TCP/UDP 23 should be blocked at the firewall, and internal use should be restricted.

The FTP service, which typically runs on port 21, falls into the same category as Telnet and is just as ancient. While FTP might be useful for public, anonymous file repositories, it is very unwise to use FTP with actual usernames and passwords. The discussion of plain text protocols above is relevant here as well. Consider replacing this service with SSH and using SCP or SFTP for your file transfers.

Port 161 UDP is the SNMP (Simple Network Management Protocol) service. You may or may not find this service running on your system depending on the brand and purpose, though it is very common to find it on routers, switches and printers. The danger here is that the service can be used to poll the system for a large variety of configuration information and may even be used to reconfigure the system. To protect against this, SNMP provides for password authentication in the form of community strings. Except for the newer forms of SNMP, all of these strings are sent in plain text across the network. Additionally, SNMP servers typically have no capability to limit requests after a predetermined number of failed attempts. This means that leaving these ports open to all provides yet another opportunity for brute force guessing of passwords. Please note that this is much easier than username and password guessing since there is no username; thus the complexity of the problem is vastly simplified.

Port 514 UDP is usually a syslog service. This service allows for remote systems to send log messages into the local system, adding them to the local syslog. While this is a great system, there is no validation of the messages. Additionally, since the service runs over UDP, it is trivial to spoof messages to be added to the syslog service. An old attacker trick is to either inject misleading messages into your system log or to simply inject so many messages that your disk fills, preventing you from logging the actual attack that may follow. Network level filtering of port 514 goes a long way to limiting where the damage can come from, but another solution might be to upgrade to SyslogNG. This package allows you to run syslog over TCP and even allows you to add authentication and encryption capabilities, preventing both of these attacks.

(3) Operating System Independent

TCP ports 20, 21, 22 and 23, while more typically found in UNIX environments, are also included in our "Independent" section. Within this group, FTP (ports 20 and 21) are most commonly found on servers (whether Windows- or UNIX-based), while ports 22 and 23 (SSH and Telnet) are found on the majority of network available devices. For this reason, all of the same issues discussed in the UNIX section exist here, but possible with greater risk. While the brute force or sniffing risks still exist, it is far less likely that any effort will be put into patching your network attached printer running a telnet service. The same goes for a switch that is running either SSH or telnet (or both) deep within our network. In the case of the printer, patches for the device simply may not be available. For the switch, patches might be overlooked if these are not "core" devices; the routers would tend to get all of our attention.

Similarly, UDP ports 161 and 514 are vulnerable for UNIX and system-wide networks. Most network-attached devices provide for some sort of SNMP management that exposes our networks and devices to, at the very least, pernicious querying of these devices for information. In a worse case, this service may be used to guess the private community string or password that can be used to reconfigure the device or query sensitive information from the device.

TCP port 79 is an oldie but goodie. This would have formerly fallen into our UNIX section, though most modern UNIX systems disable the finger service by default. The service allows you to remotely query a system to discover who is currently logged in and where they are logged in from. While this service is far from ubiquitous, it still turns up enough to be mentioned as a potential risk. Over time, many network attached devices took their service cues from UNIX. This means that many devices, routers as an example, implement the finger service. There are at least two dangers here. The obvious danger is that if the service is running, malicious individuals could potentially gather valid user IDs for the systems in question. Connected with this is the potential for a trust relationship between devices or at the very least a high probability that your user ID on one box is the same as your user ID on another box, solving half of the username/password guessing equation. The second less obvious issue is that unnecessary services tend to go unpatched. Should a vulnerability be discovered in the finger service for whatever device it is running on, we have an instant vulnerability. This service can sometimes be found running on routers, switches and RIPs (Rasterizing Image Processors, often embedded or external UNIX systems connected to color laser copiers and other network attached copiers).

TCP port 515 is commonly used as the UNIX-line printer daemon. For the moment, however, let's abstract this from port 515 and make this more generally network printer ports, which can be varied. Most business and academic environments are rife with network attached printers. This proves to be the most economical way to deploy higher-end printers to improve overall productivity and maintain an acceptable level of usability. In order for these printers to work, however, they will naturally need some sort of print server running internally. As an aside, some of these printers may be configured with an external print server, whether this is a network adapter or, say, a Windows print server. Either way, there is still a network queue available somewhere for submitting jobs. Returning to our thread, providing these kinds of services requires a significant level of computational power on the part of the printer or device managing the queue. For this reason, these devices can represent significant points of vulnerability. Why so? To abstract the problem for just a moment, let's consider the Linksys WRT-G WiFi box. This useful little wireless hub acts more or less as a wireless bridge or router, depending on how we configure it. To support this there are management interfaces available over the network (most using HTTP), and an embedded operating system. One of the most interesting developments with regard to this box was the creation of a flashable open source version of the Unix operating system with full. In fact, at a recent hacker convention, one of the talks discussed how to turn your WRT-G into a penetration testing platform! While we're not aware of any publicly available distributions to perform the same tasks on a printer, printers these days also have flash based operating systems installed. If someone were to compromise your network switch and start collecting traffic with a sniffer we would have an exposure, but at least the attacker would have to reassemble everything into meaningful information. If the printer were to be compromised we might even have a larger problem since the data arrives already assembled and ready to print!

TCP port 1080 is typically used for the SOCKS proxy service. As we did with port 515, we wish to abstract this port out to all network proxies. Now these proxies can commonly be found on ports 80, 1080, 3128, 8080 and even others, but the port number is not what's most important. The real issue is how these systems are configured.

The purpose of a proxy is typically to allow someone behind the proxy to connect to something outside of the proxy with the proxy doing all of the actual connecting, perhaps even inspecting the data that is returned. Some use proxies to attempt to anonymize their activities, concealing their actual IP address and perhaps their browser type. The danger arises with misconfigured proxies. If the proxy is not configured correctly, the proxy might be what is called an "open proxy," where anyone on the Internet can connect to a proxy and use that proxy to connect anywhere else. The worst kind of misconfiguration for a proxy occurs when the proxy allows someone outside of the proxy to connect to internal machines through the proxy. If the proxy is acting is serving as a security device or firewall, attackers would not normally be able to directly probe our networks. With the proxy misconfigured, however, an attacker could use the proxy itself to perform probes, in essence turning our proxy server against us.


While we've just scratched the surface of port vulnerabilities, the examples above should be convincing for even the risk-tolerant that continuous oversight of their firewalls and routers are in order.



Additional detail on TCP/IP ports may be found in several places. The definitive site is that of the Internet Assigned Numbers Authority (IANA) at The official port assignments may be found therein at Two derivative Web sites provide an easy-to-use interface to the IANA list: the Internet Ports Database at, and our own Internet Ports Pass at

Arguably best, reliable source of information on port vulnerabilities is the SANS Internet Storm Center at This site is an exceedingly rich repository of reports, charts, databases, statistics and archives on all aspects of port vulnerabilities. The port lookup feature (top-left) provides ready access to port-specific vulnerabilities. The "Top 20" list is also a valuable resource (

For those who really want to get serious about port security, we recommend . The page says it all: "Snort is the de facto standard for intrusion detection/prevention."

For a refresher on packet-based networks, including a discussion of appliances and ports, the short animated film Warriors of the Net ( is entertaining and well worth the twelve minutes invested.

Hal Berghel is an educator, administrator, inventor, author, columnist, lecturer and sometimes talk show guest. He is both an ACM and IEEE Fellow and has been recognized by both organizations for distinguished service. He is the Associate Dean of the Howard R. Hughes College of Engineering at UNLV, and his consultancy, Berghel.Net, provides security services for government and industry.

David Hoelzer is a Director of the Internet Forensics Lab at CCR and a faculty member of the SANS Institute.  He is the owner of CyberDefense (, a computer security and forensics consultancy.